starlingx
/
ha
Code Issues Proposed changes

Merge remote-tracking branch 'origin/master' into f/centos8 

Signed-off-by: Charles Short <charles.short@windriver.com>
Change-Id: I59d1e0fae5c3767febdb2f2c5eaab731c7279268
This commit is contained in:
Charles Short 2021-05-19 15:33:52 -04:00
parent c3b94d1f2f 74fda90dac
commit 85bab5d2b3
52 changed files with 980 additions and 274 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
.zuul.yaml
View File

@ -4,14 +4,17 @@
- publish-stx-docs
- stx-api-ref-jobs
- stx-release-notes-jobs
- stx-bandit-jobs
check:
jobs:
- openstack-tox-pep8
- openstack-tox-linters
- ha-tox-pylint
gate:
jobs:
- openstack-tox-pep8
- openstack-tox-linters
- ha-tox-pylint
post:
jobs:
- stx-ha-upload-git-mirror
@ -55,6 +58,14 @@
nfv: https://opendev.org/starlingx/nfv
update: https://opendev.org/starlingx/update
- job:
name: ha-tox-pylint
parent: tox
description: |
Run pylint test for ha in python2.7
nodeset: ubuntu-bionic
vars:
tox_envlist: pylint
- job:
name: stx-ha-upload-git-mirror
parent: upload-git-mirror

10
api-ref/source/conf.py
View File

@ -50,10 +50,12 @@ source_suffix = '.rst'
master_doc = 'index'
# General information about the project.
repository_name = 'openstack/stx-ha'
project = u'StarlingX HA'
bug_project = 'starlingx'
bug_tag = 'stx.ha'
# openstackdocstheme options
openstackdocs_repo_name = 'starlingx/ha'
openstackdocs_use_storyboard = True
openstackdocs_auto_name = False
# If true, the current module name will be prepended to all description
# unit titles (such as .. function::).
@ -64,7 +66,7 @@ add_module_names = False
show_authors = False
# The name of the Pygments (syntax highlighting) style to use.
pygments_style = 'sphinx'
pygments_style = 'native'
# Grouping the document tree into LaTeX files. List of tuples
# (source start file, target name, title, author, documentclass

6
doc/requirements.txt
View File

@ -1,8 +1,8 @@
sphinx>=1.6.2
openstackdocstheme>=1.26.0 # Apache-2.0
sphinx>=2.0.0,!=2.1.0 # BSD
openstackdocstheme>=2.2.1 # Apache-2.0
# Release Notes
reno>=0.1.1 # Apache-2.0
reno>=3.1.0 # Apache-2.0
# API Reference Guide
os-api-ref>=1.4.0 # Apache-2.0

13
doc/source/conf.py
View File

@ -25,11 +25,6 @@ project = u'StarlingX HA'
copyright = u'2018, StarlingX'
author = u'StarlingX'
# The short X.Y version
version = u''
# The full version, including alpha/beta/rc tags
release = u'0.1'
# -- General configuration ---------------------------------------------------
@ -56,8 +51,10 @@ source_suffix = '.rst'
# The master toctree document.
master_doc = 'index'
bug_project = 'starlingx'
bug_tag = 'stx.ha'
# openstackdocstheme options
openstackdocs_repo_name = 'starlingx/ha'
openstackdocs_use_storyboard = True
openstackdocs_auto_name = False
# The language for content autogenerated by Sphinx. Refer to documentation
# for a list of supported languages.
@ -72,7 +69,7 @@ language = None
exclude_patterns = []
# The name of the Pygments (syntax highlighting) style to use.
pygments_style = 'sphinx'
pygments_style = 'native'
# -- Options for HTML output -------------------------------------------------

12
releasenotes/source/conf.py
View File

@ -41,8 +41,10 @@ extensions = [
'reno.sphinxext',
]
bug_project = 'starlingx'
bug_tag = 'stx.ha'
# openstackdocstheme options
openstackdocs_repo_name = 'starlingx/ha'
openstackdocs_use_storyboard = True
openstackdocs_auto_name = False
# Add any paths that contain templates here, relative to this directory.
# templates_path = ['_templates']
@ -93,7 +95,7 @@ exclude_patterns = []
# show_authors = False
# The name of the Pygments (syntax highlighting) style to use.
pygments_style = 'sphinx'
pygments_style = 'native'
# A list of ignored prefixes for module index sorting.
# modindex_common_prefix = []
@ -142,10 +144,6 @@ html_theme = 'starlingxdocs'
# directly to the root of the documentation.
# html_extra_path = []
# If not '', a 'Last updated on:' timestamp is inserted at every page bottom,
# using the given strftime format.
html_last_updated_fmt = '%Y-%m-%d %H:%M'
# If true, SmartyPants will be used to convert quotes and dashes to
# typographically correct entities.
# html_use_smartypants = True

2
service-mgmt-api/centos/build_srpm.data
View File

@ -1,4 +1,4 @@
SRC_DIR=sm-api
TAR_NAME=sm-api
VERSION=1.0
TIS_PATCH_VER=4
TIS_PATCH_VER=PKG_GITREVCOUNT

2
service-mgmt-client/centos/build_srpm.data
View File

@ -1,4 +1,4 @@
SRC_DIR=sm-client
TAR_NAME=sm-client
VERSION=1.0
TIS_PATCH_VER=2
TIS_PATCH_VER=PKG_GITREVCOUNT

2
service-mgmt-client/sm-client/sm_client/shell.py
View File

@ -1,5 +1,5 @@
#
# Copyright (c) 2014-2018 Wind River Systems, Inc.
# Copyright (c) 2014-2020 Wind River Systems, Inc.
#
# SPDX-License-Identifier: Apache-2.0
#

2
service-mgmt-tools/centos/build_srpm.data
View File

@ -1,4 +1,4 @@
SRC_DIR=sm-tools
TAR_NAME=sm-tools
VERSION=1.0
TIS_PATCH_VER=2
TIS_PATCH_VER=PKG_GITREVCOUNT

2
service-mgmt/sm-common/centos/build_srpm.data
View File

@ -1,2 +1,2 @@
SRC_DIR=$PKG_BASE
TIS_PATCH_VER=20
TIS_PATCH_VER=PKG_GITREVCOUNT+15

5
service-mgmt/sm-common/src/sm_hw.c
View File

@ -232,6 +232,11 @@ SmErrorT sm_hw_get_if_state( const char if_name[], bool* enabled )
struct ifreq if_data;
SmHwThreadInfoT* thread_info;
if(strcmp(if_name, "lo") == 0)
{
*enabled = true;
return SM_OKAY;
}
thread_info = sm_hw_find_thread_info();
if( NULL == thread_info )
{

16
service-mgmt/sm-common/src/sm_node_utils.c
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2014-2018 Wind River Systems, Inc.
// Copyright (c) 2014-2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
@ -584,3 +584,17 @@ SmErrorT sm_node_utils_is_aio_duplex( bool* is_aio_duplex )
return SM_OKAY;
}
// ****************************************************************************
// Node Utilities - Clear the unhealthy flag
// ==============================
extern void sm_node_utils_reset_unhealthy_flag( void )
{
if( 0 == access( SM_NODE_UNHEALTHY_FILE, F_OK ) )
{
unlink( SM_NODE_UNHEALTHY_FILE );
if( 0 == access( SM_NODE_UNHEALTHY_FILE, F_OK ) )
{
DPRINTFE("file did not get removed ; %s", SM_NODE_UNHEALTHY_FILE);
}
}
}

8
service-mgmt/sm-common/src/sm_node_utils.h
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2014 Wind River Systems, Inc.
// Copyright (c) 2014,2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
@ -104,6 +104,12 @@ extern SmErrorT sm_node_utils_is_aio_duplex( bool* is_aio_duplex );
extern bool sm_node_utils_set_failover( bool to_disable );
// ****************************************************************************
// ****************************************************************************
// Node Utilities - Clear the unhealthy flag
// ==============================
extern void sm_node_utils_reset_unhealthy_flag( void );
// ****************************************************************************
#ifdef __cplusplus
}
#endif

23
service-mgmt/sm-common/src/sm_types.c
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2014-2018 Wind River Systems, Inc.
// Copyright (c) 2014-2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
@ -173,6 +173,7 @@ _sm_failover_event_mappings[SM_FAILOVER_EVENT_MAX] =
{SM_FAILOVER_EVENT_HEARTBEAT_ENABLED, "heartbeat-enabled"},
{SM_FAILOVER_EVENT_IF_STATE_CHANGED, "interface-state-changed"},
{SM_FAILOVER_EVENT_FAIL_PENDING_TIMEOUT, "fail-pending-timeout"},
{SM_FAILOVER_EVENT_FAILED_RECOVERY_AUDIT, "failed-recovery-audit"},
{SM_FAILOVER_EVENT_NODE_ENABLED, "node-enabled"}
};
@ -186,6 +187,15 @@ _sm_failover_state_mappings[SM_FAILOVER_STATE_MAX] =
{SM_FAILOVER_STATE_SURVIVED, "survived"}
};
static SmValueStrMappingT
_sm_failover_interface_state_mappings[SM_FAILOVER_INTERFACE_STATE_MAX] =
{
{SM_FAILOVER_INTERFACE_UNKNOWN, "unknown"},
{SM_FAILOVER_INTERFACE_OK, "ok"},
{SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT, "missing-heartbeat"},
{SM_FAILOVER_INTERFACE_DOWN, "down"}
};
static SmValueStrMappingT
_sm_service_domain_neighbor_state_mappings[SM_SERVICE_DOMAIN_NEIGHBOR_STATE_MAX] =
{
@ -993,6 +1003,17 @@ const char* sm_failover_state_str( SmFailoverStateT state )
SM_FAILOVER_STATE_MAX,
state ) );
}
// ****************************************************************************
// Types - Failover Interface State String
// =============================================
const char* sm_failover_interface_state_str( SmFailoverInterfaceStateT state )
{
return( sm_mapping_get_str( _sm_failover_interface_state_mappings,
SM_FAILOVER_INTERFACE_STATE_MAX,
state ) );
}
// ****************************************************************************
// ****************************************************************************

10
service-mgmt/sm-common/src/sm_types.h
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2014-2018 Wind River Systems, Inc.
// Copyright (c) 2014-2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
@ -293,6 +293,7 @@ typedef enum{
SM_FAILOVER_EVENT_HEARTBEAT_ENABLED,
SM_FAILOVER_EVENT_IF_STATE_CHANGED,
SM_FAILOVER_EVENT_FAIL_PENDING_TIMEOUT,
SM_FAILOVER_EVENT_FAILED_RECOVERY_AUDIT,
SM_FAILOVER_EVENT_NODE_ENABLED,
SM_FAILOVER_EVENT_MAX
}SmFailoverEventT;
@ -730,7 +731,8 @@ typedef enum
SM_FAILOVER_INTERFACE_UNKNOWN,
SM_FAILOVER_INTERFACE_OK,
SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT,
SM_FAILOVER_INTERFACE_DOWN
SM_FAILOVER_INTERFACE_DOWN,
SM_FAILOVER_INTERFACE_STATE_MAX
}SmFailoverInterfaceStateT;
// ****************************************************************************
@ -960,6 +962,10 @@ extern const char* sm_failover_event_str( SmFailoverEventT event );
extern const char* sm_failover_state_str( SmFailoverStateT state );
// ****************************************************************************
const char* sm_failover_interface_state_str( SmFailoverInterfaceStateT state );
// ****************************************************************************
// ****************************************************************************
// Types - Service Domain Neighbor State Value
// ===========================================

2
service-mgmt/sm-db/centos/build_srpm.data
View File

@ -2,4 +2,4 @@ SRC_DIR=`pwd`
COPY_LIST="$PKG_BASE/LICENSE"
TAR_NAME=sm-db
VERSION=1.0.0
TIS_PATCH_VER=31
TIS_PATCH_VER=PKG_GITREVCOUNT+19

93
service-mgmt/sm-db/database/create_sm_db.sql
View File

@ -52,7 +52,6 @@ INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(20,'yes','controller-services','mtc-a
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(22,'yes','controller-services','hw-mon','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(23,'yes','controller-services','dnsmasq','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(24,'yes','controller-services','fm-mgr','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(25,'yes','controller-services','snmp','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(26,'yes','cloud-services','keystone','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(49,'yes','directory-services','open-ldap','critical');
INSERT INTO "SERVICE_GROUP_MEMBERS" VALUES(50,'yes','web-services','lighttpd','critical');
@ -95,6 +94,8 @@ INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'yes','controller-service
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','distributed-cloud-services','dcdbsync-api','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','ironic-ip','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','distributed-cloud-services','dcdbsync-openstack-api','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','distributed-cloud-services','dcmanager-orchestrator','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','distributed-cloud-services','dcmanager-audit-worker','critical' FROM "SERVICE_GROUP_MEMBERS";
CREATE TABLE SERVICES ( ID INTEGER PRIMARY KEY AUTOINCREMENT, PROVISIONED CHAR(32), NAME CHAR(32), DESIRED_STATE CHAR(32), STATE CHAR(32), STATUS CHAR(32), CONDITION CHAR(32), MAX_FAILURES INT, FAIL_COUNTDOWN INT, FAIL_COUNTDOWN_INTERVAL INT, MAX_ACTION_FAILURES INT, MAX_TRANSITION_FAILURES INT, PID_FILE CHAR(256) );
INSERT INTO "SERVICES" VALUES(1,'yes','oam-ip','initial','initial','none','none',2,1,90000,4,16,'');
INSERT INTO "SERVICES" VALUES(2,'yes','management-ip','initial','initial','none','none',2,1,90000,4,16,'');
@ -117,7 +118,6 @@ INSERT INTO "SERVICES" VALUES(23,'yes','dnsmasq','initial','initial','none','non
INSERT INTO "SERVICES" VALUES(24,'yes','fm-mgr','initial','initial','none','none',2,1,90000,4,16,'/var/run/fmManager.pid');
INSERT INTO "SERVICES" VALUES(25,'yes','keystone','initial','initial','none','none',2,1,90000,4,16,'/var/run/openstack-keystone.pid');
INSERT INTO "SERVICES" VALUES(48,'yes','open-ldap','initial','initial','none','none',2,1,90000,4,16,'/var/run/slapd.pid');
INSERT INTO "SERVICES" VALUES(49,'yes','snmp','initial','initial','none','none',32,16,90000,4,16,'/var/run/snmpd.pid');
INSERT INTO "SERVICES" VALUES(50,'yes','lighttpd','initial','initial','none','none',2,1,90000,4,16,'/var/run/lighttpd.pid');
INSERT INTO "SERVICES" VALUES(51,'yes','horizon','initial','initial','none','none',2,1,90000,4,16,'/var/run/openstack-dashboard.pid');
INSERT INTO "SERVICES" VALUES(52,'yes','patch-alarm-manager','initial','initial','none','none',2,1,90000,4,16,'/var/run/patch-alarm-manager.pid');
@ -158,6 +158,8 @@ INSERT INTO "SERVICES" SELECT MAX(id) + 1,'yes','cluster-host-ip','initial','ini
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','dcdbsync-api','initial','initial','none','none',2,1,90000,4,16,'/var/run/resource-agents/dcdbsync-api.pid' FROM "SERVICES";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','ironic-ip','initial','initial','none','none',2,1,90000,4,16,'' FROM "SERVICES";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','dcdbsync-openstack-api','initial','initial','none','none',2,1,90000,4,16,'/var/run/resource-agents/dcdbsync-openstack-api.pid' FROM "SERVICES";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','dcmanager-orchestrator','initial','initial','none','none',2,1,90000,4,16,'/var/run/resource-agents/dcmanager-orchestrator.pid' FROM "SERVICES";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','dcmanager-audit-worker','initial','initial','none','none',2,1,90000,4,16,'/var/run/resource-agents/dcmanager-audit-worker.pid' FROM "SERVICES";
CREATE TABLE SERVICE_HEARTBEAT ( ID INTEGER PRIMARY KEY AUTOINCREMENT, PROVISIONED CHAR(32), NAME CHAR(32), TYPE CHAR(32), SRC_ADDRESS CHAR(108), SRC_PORT INT, DST_ADDRESS CHAR(108), DST_PORT INT, MESSAGE CHAR(256), INTERVAL_IN_MS INT, MISSED_WARN INT, MISSED_DEGRADE INT, MISSED_FAIL INT, STATE CHAR(32), MISSED INT, HEARTBEAT_TIMER_ID INT, HEARTBEAT_SOCKET INT );
CREATE TABLE SERVICE_DEPENDENCY ( DEPENDENCY_TYPE CHAR(32), SERVICE_NAME CHAR(32), STATE CHAR(32), ACTION CHAR(32), DEPENDENT CHAR(32), DEPENDENT_STATE CHAR(32), PRIMARY KEY (DEPENDENCY_TYPE, SERVICE_NAME, STATE, ACTION, DEPENDENT));
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','oam-ip','not-applicable','enable','management-ip','enabled-active');
@ -188,9 +190,6 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','vim-api','not-applicable','ena
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','vim-webserver','not-applicable','enable','vim','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dnsmasq','not-applicable','enable','sysinv-conductor','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','fm-mgr','not-applicable','enable','postgres','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','snmp','not-applicable','enable','oam-ip','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','snmp','not-applicable','enable','postgres','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','snmp','not-applicable','enable','platform-nfs-ip','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','etcd','not-applicable','enable','etcd-fs','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','keystone','not-applicable','enable','postgres','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','keystone','not-applicable','enable','rabbit','enabled-active');
@ -206,16 +205,19 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','management-ip','not-applicable
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','management-ip','not-applicable','disable','rabbit-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','management-ip','not-applicable','disable','platform-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-pg','not-applicable','go-standby','pg-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-pg','not-applicable','disable','pg-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-rabbit','not-applicable','go-standby','rabbit-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-rabbit','not-applicable','disable','rabbit-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-platform','not-applicable','go-standby','platform-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-platform','not-applicable','disable','platform-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-extension','not-applicable','go-standby','extension-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-extension','not-applicable','disable','extension-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','pg-fs','not-applicable','disable','postgres','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rabbit-fs','not-applicable','disable','rabbit','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','nfs-mgmt','not-applicable','disable','platform-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','extension-fs','not-applicable','disable','extension-export-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','platform-nfs-ip','not-applicable','disable','keystone','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','platform-fs','not-applicable','disable','platform-export-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','postgres','not-applicable','disable','snmp','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','postgres','not-applicable','disable','fm-mgr','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','postgres','not-applicable','disable','keystone','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','postgres','not-applicable','disable','sysinv-conductor','disabled');
@ -239,8 +241,8 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-platform','not-applicable
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','etcd-fs','not-applicable','disable','etcd','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cinder-lvm','not-applicable','disable','iscsi','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-cinder','not-applicable','go-standby','cinder-lvm','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-cinder','not-applicable','disable','cinder-lvm','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','ceph-manager','not-applicable','disable','sysinv-conductor','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','snmp','not-applicable','enable','dnsmasq','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-engine','not-applicable','enable','rabbit','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-engine','not-applicable','enable','postgres','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-engine','not-applicable','enable','management-ip','enabled-active');
@ -259,8 +261,10 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','postgres','not-applicable','di
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','management-ip','not-applicable','disable','dcmanager-manager','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-api','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','dcmanager-api','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-audit','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','dcmanager-audit','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-audit','not-applicable','enable','dcmanager-audit-worker','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-audit-worker','not-applicable','disable','dcmanager-audit','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-orchestrator','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','dcmanager-orchestrator','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-sysinv-api-proxy','not-applicable','enable','sysinv-inv','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-sysinv-api-proxy','not-applicable','enable','dcorch-engine','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','sysinv-inv','not-applicable','disable','dcorch-sysinv-api-proxy','disabled');
@ -278,6 +282,7 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-engine','not-applicable
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-dc-vault','not-applicable','go-active','management-ip','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dc-vault-fs','not-applicable','enable','drbd-dc-vault','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-dc-vault','not-applicable','go-standby','dc-vault-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-dc-vault','not-applicable','disable','dc-vault-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcorch-patch-api-proxy','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','dcorch-patch-api-proxy','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-etcd','not-applicable','go-active','management-ip','enabled-active');
@ -295,10 +300,11 @@ INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcdbsync-api','not-applicable'
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','keystone','not-applicable','disable','dcdbsync-api','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcdbsync-openstack-api','not-applicable','enable','keystone','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','keystone','not-applicable','disable','dcdbsync-openstack-api','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-audit-worker','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','dcmanager-audit-worker','disabled');
CREATE TABLE SERVICE_INSTANCES ( ID INTEGER PRIMARY KEY AUTOINCREMENT, SERVICE_NAME CHAR(32), INSTANCE_NAME CHAR(32), INSTANCE_PARAMETERS CHAR(1024) );
INSERT INTO "SERVICE_INSTANCES" VALUES(1,'lighttpd','lighttpd','');
INSERT INTO "SERVICE_INSTANCES" VALUES(2,'horizon','horizon','');
INSERT INTO "SERVICE_INSTANCES" VALUES(3,'snmp','snmp','');
INSERT INTO "SERVICE_INSTANCES" VALUES(4,'patch-alarm-manager','patch-alarm-manager','');
INSERT INTO "SERVICE_INSTANCES" VALUES(5,'hw-mon','hw-mon','');
INSERT INTO "SERVICE_INSTANCES" VALUES(6,'guest-agent','guest-agent','');
@ -421,10 +427,6 @@ INSERT INTO "SERVICE_ACTIONS" VALUES('open-ldap','enable','lsb-script','','openl
INSERT INTO "SERVICE_ACTIONS" VALUES('open-ldap','disable','lsb-script','','openldap','stop','',1,1,1,15,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('open-ldap','audit-enabled','lsb-script','','openldap','status','',2,2,2,15,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('open-ldap','audit-disabled','lsb-script','','openldap','status','',0,0,0,15,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('snmp','enable','lsb-script','','snmpd','start','',2,2,2,15,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('snmp','disable','lsb-script','','snmpd','stop','',1,1,1,15,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('snmp','audit-enabled','lsb-script','','snmpd','status','',2,2,2,15,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('snmp','audit-disabled','lsb-script','','snmpd','status','',0,0,0,15,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('lighttpd','enable','lsb-script','','lighttpd','start','',2,2,2,15,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('lighttpd','disable','lsb-script','','lighttpd','stop','',1,1,1,15,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('lighttpd','audit-enabled','lsb-script','','lighttpd','status','',2,2,2,15,40);
@ -499,6 +501,10 @@ INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit','enable','ocf-script','op
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit','disable','ocf-script','openstack','dcmanager-audit','stop','',1,1,1,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit','audit-enabled','ocf-script','openstack','dcmanager-audit','monitor','',2,2,2,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit','audit-disabled','ocf-script','openstack','dcmanager-audit','monitor','',0,0,0,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-orchestrator','enable','ocf-script','openstack','dcmanager-orchestrator','start','',2,2,2,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-orchestrator','disable','ocf-script','openstack','dcmanager-orchestrator','stop','',1,1,1,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-orchestrator','audit-enabled','ocf-script','openstack','dcmanager-orchestrator','monitor','',2,2,2,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-orchestrator','audit-disabled','ocf-script','openstack','dcmanager-orchestrator','monitor','',0,0,0,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-api','enable','ocf-script','openstack','dcmanager-api','start','',2,2,2,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-api','disable','ocf-script','openstack','dcmanager-api','stop','',1,1,1,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-api','audit-enabled','ocf-script','openstack','dcmanager-api','monitor','',2,2,2,20,5);
@ -572,6 +578,10 @@ INSERT INTO "SERVICE_ACTIONS" VALUES('ironic-ip','enable','ocf-script','heartbea
INSERT INTO "SERVICE_ACTIONS" VALUES('ironic-ip','disable','ocf-script','heartbeat','IPaddr2','stop','',1,1,1,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('ironic-ip','audit-enabled','ocf-script','heartbeat','IPaddr2','monitor','',2,2,2,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('ironic-ip','audit-disabled','ocf-script','heartbeat','IPaddr2','monitor','',0,0,0,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit-worker','enable','ocf-script','openstack','dcmanager-audit-worker','start','',2,2,2,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit-worker','disable','ocf-script','openstack','dcmanager-audit-worker','stop','',1,1,1,20,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit-worker','audit-enabled','ocf-script','openstack','dcmanager-audit-worker','monitor','',2,2,2,20,5);
INSERT INTO "SERVICE_ACTIONS" VALUES('dcmanager-audit-worker','audit-disabled','ocf-script','openstack','dcmanager-audit-worker','monitor','',0,0,0,20,5);
CREATE TABLE SERVICE_ACTION_RESULTS ( PLUGIN_TYPE CHAR(32), PLUGIN_NAME CHAR(80), PLUGIN_COMMAND CHAR(80), PLUGIN_EXIT_CODE CHAR(32), ACTION_RESULT CHAR(32), SERVICE_STATE CHAR(32), SERVICE_STATUS CHAR(32), SERVICE_CONDITION CHAR(32), PRIMARY KEY (PLUGIN_TYPE, PLUGIN_NAME, PLUGIN_COMMAND, PLUGIN_EXIT_CODE));
INSERT INTO "SERVICE_ACTION_RESULTS" VALUES('lsb-script','default','status','0','success','enabled-active','unknown','unknown');
INSERT INTO "SERVICE_ACTION_RESULTS" VALUES('lsb-script','default','status','1','success','disabled','unknown','unknown');
@ -642,6 +652,7 @@ INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','dockerdistribution-fs','initial'
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dockerdistribution-fs','not-applicable','enable','drbd-dockerdistribution','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','docker-distribution','not-applicable','enable','dockerdistribution-fs','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-dockerdistribution','not-applicable','go-standby','dockerdistribution-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-dockerdistribution','not-applicable','disable','dockerdistribution-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dockerdistribution-fs','not-applicable','disable','docker-distribution','disabled');
INSERT INTO "SERVICE_ACTIONS" VALUES('dockerdistribution-fs','enable','ocf-script','heartbeat','Filesystem','start','',2,2,2,60,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('dockerdistribution-fs','disable','ocf-script','heartbeat','Filesystem','stop','',1,1,1,180,'');
@ -656,6 +667,38 @@ INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-dockerdistribution','go-standby','ocf
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-dockerdistribution','audit-enabled','ocf-script','linbit','drbd','monitor','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',2,2,2,20,30);
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-dockerdistribution','audit-disabled','ocf-script','linbit','drbd','monitor','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',0,0,0,20,28);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','rook-mon-exit','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1, 'no','rook-mon-exit','initial','initial','none','none',2,1,90000,4,16,'' FROM "SERVICES";
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rook-mon-exit','not-applicable','enable','docker-distribution','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','docker-distribution','not-applicable','disable','rook-mon-exit','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rook-mon-exit','not-applicable','enable','cluster-host-ip','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cluster-host-ip','not-applicable','disable','rook-mon-exit','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rook-mon-exit','not-applicable','enable','etcd','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','etcd','not-applicable','disable','rook-mon-exit','disabled');
INSERT INTO "SERVICE_ACTIONS" VALUES('rook-mon-exit','enable','lsb-script','','rook-mon-exit','start','',2,2,2,30,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('rook-mon-exit','disable','lsb-script','','rook-mon-exit','stop','',0,0,0,360,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('rook-mon-exit','audit-enabled','lsb-script','','rook-mon-exit','status','',2,2,2,15,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('rook-mon-exit','audit-disabled','lsb-script','','rook-mon-exit','status','',0,0,0,15,40);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','rookmon-fs','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','rookmon-fs','initial','initial','none','none',2,1,90000,4,16,'' FROM "SERVICES";
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rookmon-fs','not-applicable','enable','drbd-rookmon','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rook-mon-exit','not-applicable','enable','rookmon-fs','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-rookmon','not-applicable','go-standby','rookmon-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-rookmon','not-applicable','disable','rookmon-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','rookmon-fs','not-applicable','disable','rook-mon-exit','disabled');
INSERT INTO "SERVICE_ACTIONS" VALUES('rookmon-fs','enable','ocf-script','heartbeat','Filesystem','start','',2,2,2,60,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('rookmon-fs','disable','ocf-script','heartbeat','Filesystem','stop','',1,1,1,180,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('rookmon-fs','audit-enabled','ocf-script','heartbeat','Filesystem','monitor','',2,2,2,60,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('rookmon-fs','audit-disabled','ocf-script','heartbeat','Filesystem','monitor','',0,0,0,60,40);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','drbd-rookmon','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','drbd-rookmon','initial','initial','none','none',2,1,90000,4,16,'' FROM "SERVICES";
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','enable','ocf-script','linbit','drbd','start','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',2,2,2,90,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','disable','ocf-script','linbit','drbd','stop','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',1,1,1,180,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','go-active','ocf-script','linbit','drbd','promote','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',2,2,2,180,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','go-standby','ocf-script','linbit','drbd','demote','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',2,2,2,180,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','audit-enabled','ocf-script','linbit','drbd','monitor','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',2,2,2,20,30);
INSERT INTO "SERVICE_ACTIONS" VALUES('drbd-rookmon','audit-disabled','ocf-script','linbit','drbd','monitor','master_max=1,master_node_max=1,clone_max=2,clone_node_max=1,notify=true,globally_unique=false',0,0,0,20,28);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','ceph-mon','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1, 'no','ceph-mon','initial','initial','none','none',2,1,90000,4,16,'/var/run/ceph/mon.controller.pid' FROM "SERVICES";
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','ceph-mon','not-applicable','enable','management-ip','enabled-active');
@ -669,6 +712,7 @@ INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','cephmon-fs','initial','initial',
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cephmon-fs','not-applicable','enable','drbd-cephmon','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','ceph-mon','not-applicable','enable','cephmon-fs','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-cephmon','not-applicable','go-standby','cephmon-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','drbd-cephmon','not-applicable','disable','cephmon-fs','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cephmon-fs','not-applicable','disable','ceph-mon','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','ceph-radosgw','not-applicable','enable','ceph-mon','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','ceph-mon','not-applicable','disable','ceph-radosgw','disabled');
@ -735,4 +779,25 @@ INSERT INTO "SERVICE_ACTIONS" VALUES('dc-iso-fs','disable','ocf-script','heartbe
INSERT INTO "SERVICE_ACTIONS" VALUES('dc-iso-fs','audit-enabled','ocf-script','heartbeat','Filesystem','monitor','',2,2,2,30,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('dc-iso-fs','audit-disabled','ocf-script','heartbeat','Filesystem','monitor','',0,0,0,30,40);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','cert-mon','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','cert-mon','initial','initial','none','none',2,1,90000,4,16,'/var/run/cert-mon.pid' FROM "SERVICES";
INSERT INTO "SERVICE_INSTANCES" SELECT MAX(id) + 1,'cert-mon','cert-mon','' FROM "SERVICE_INSTANCES";
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cert-mon','not-applicable','enable','sysinv-inv','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','sysinv-inv','not-applicable','disable','cert-mon','disabled');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','cert-mon','not-applicable','enable','dcmanager-manager','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','dcmanager-manager','not-applicable','disable','cert-mon','disabled');
INSERT INTO "SERVICE_ACTIONS" VALUES('cert-mon','enable','ocf-script','platform','cert-mon','start','',2,2,2,30,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('cert-mon','disable','ocf-script','platform','cert-mon','stop','',1,1,1,60,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('cert-mon','audit-enabled','ocf-script','platform','cert-mon','monitor','',2,2,2,30,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('cert-mon','audit-disabled','ocf-script','platform','cert-mon','monitor','',0,0,0,30,40);
INSERT INTO "SERVICE_GROUP_MEMBERS" SELECT MAX(id) + 1,'no','controller-services','device-image-fs','critical' FROM "SERVICE_GROUP_MEMBERS";
INSERT INTO "SERVICES" SELECT MAX(id) + 1,'no','device-image-fs','initial','initial','none','none',2,1,90000,4,16,'' FROM "SERVICES";
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','device-image-fs','not-applicable','enable','platform-fs','enabled-active');
INSERT INTO "SERVICE_DEPENDENCY" VALUES('action','platform-fs','not-applicable','disable','device-image-fs','disabled');
INSERT INTO "SERVICE_ACTIONS" VALUES('device-image-fs','enable','ocf-script','heartbeat','Filesystem','start','',2,2,2,60,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('device-image-fs','disable','ocf-script','heartbeat','Filesystem','stop','',1,1,1,180,'');
INSERT INTO "SERVICE_ACTIONS" VALUES('device-image-fs','audit-enabled','ocf-script','heartbeat','Filesystem','monitor','',2,2,2,60,40);
INSERT INTO "SERVICE_ACTIONS" VALUES('device-image-fs','audit-disabled','ocf-script','heartbeat','Filesystem','monitor','',0,0,0,60,40);
COMMIT;

17
service-mgmt/sm-db/src/sm_db.c
View File

@ -164,13 +164,18 @@ SmErrorT sm_db_statement_finalize( SmDbStatementT* sm_db_statement )
// ****************************************************************************
// Database - Connect
// ==================
SmErrorT sm_db_connect( const char* sm_db_name, SmDbHandleT** sm_db_handle )
SmErrorT sm_db_connect( const char* sm_db_name, SmDbHandleT** sm_db_handle, bool readonly )
{
int rc;
rc = sqlite3_open_v2( sm_db_name, (sqlite3**) sm_db_handle,
SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE |
SQLITE_OPEN_NOMUTEX, NULL );
int flag;
if (readonly)
{
flag = SQLITE_OPEN_READONLY | SQLITE_OPEN_NOMUTEX;
} else
{
flag = SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE | SQLITE_OPEN_NOMUTEX;
}
rc = sqlite3_open_v2( sm_db_name, (sqlite3**) sm_db_handle, flag, NULL );
if( SQLITE_OK != rc )
{
DPRINTFE( "Failed to connect to database (%s), rc=%i.", sm_db_name, rc );
@ -1128,4 +1133,4 @@ SmErrorT sm_db_finalize( void )
return( SM_OKAY );
}
// ****************************************************************************
// ****************************************************************************

2
service-mgmt/sm-db/src/sm_db.h
View File

@ -59,7 +59,7 @@ extern SmErrorT sm_db_statement_finalize( SmDbStatementT* sm_db_statement );
// Database - Connect
// ==================
extern SmErrorT sm_db_connect( const char* sm_db_name,
SmDbHandleT** sm_db_handle );
SmDbHandleT** sm_db_handle, bool readonly = false );
// ****************************************************************************
// ****************************************************************************

2
service-mgmt/sm-db/src/sm_db_iterator.c
View File

@ -123,7 +123,7 @@ SmErrorT sm_db_iterator_initialize( const char* db_name,
memset( it, 0, sizeof(SmDbIteratorT) );
error = sm_db_connect( db_name, &(it->sm_db_handle) );
error = sm_db_connect( db_name, &(it->sm_db_handle), true );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.", db_name,

2
service-mgmt/sm/centos/build_srpm.data
View File

@ -2,4 +2,4 @@ SRC_DIR=$PKG_BASE
COPY_LIST="$PKG_BASE/LICENSE"
TAR_NAME=sm
VERSION=1.0.0
TIS_PATCH_VER=33
TIS_PATCH_VER=PKG_GITREVCOUNT+24

2
service-mgmt/sm/scripts/sm.service
View File

@ -11,6 +11,8 @@ ExecStart=/etc/init.d/sm start
ExecStop=/etc/init.d/sm stop
PIDFile=/var/run/sm.pid
KillMode=process
RestartSec=10
Restart=on-failure
[Install]
WantedBy=multi-user.target

174
service-mgmt/sm/src/sm_cluster_hbs_info_msg.cpp
View File

@ -22,6 +22,7 @@
#include "sm_limits.h"
#include "sm_selobj.h"
#include "sm_worker_thread.h"
#include "sm_node_utils.h"
// uncomment when debugging this module to enabled DPRINTFD output to log file
// #define __DEBUG__MSG__
@ -44,6 +45,7 @@ static const unsigned int size_of_msg_header =
bool operator==(const SmClusterHbsInfoT& lhs, const SmClusterHbsInfoT& rhs)
{
return lhs.storage0_responding == rhs.storage0_responding &&
lhs.sm_heartbeat_fail == rhs.sm_heartbeat_fail &&
lhs.number_of_node_reachable == rhs.number_of_node_reachable;
}
@ -86,25 +88,29 @@ void log_cluster_hbs_state(const SmClusterHbsStateT& state)
if(state.storage0_enabled)
{
DPRINTFI("Cluster hbs last updated %d secs ago, storage-0 is provisioned, "
"from controller-0: %d nodes enabled, %d nodes reachable, storage-0 %s responding "
"from controller-1: %d nodes enabled, %d nodes reachable, storage-0 %s responding",
DPRINTFI("Cluster hbs last updated %d secs ago, storage-0 is provisioned,\n"
"from controller-0: SM %s, %d nodes enabled, %d nodes reachable, storage-0 %s responding\n"
"from controller-1: SM %s, %d nodes enabled, %d nodes reachable, storage-0 %s responding",
secs_since_update,
state.controllers[0].sm_heartbeat_fail ? "FAILED": "ok ",
state.controllers[0].number_of_node_enabled,
state.controllers[0].number_of_node_reachable,
state.controllers[0].storage0_responding ? "is" : "is not",
state.controllers[1].sm_heartbeat_fail ? "FAILED": "ok ",
state.controllers[1].number_of_node_enabled,
state.controllers[1].number_of_node_reachable,
state.controllers[1].storage0_responding ? "is" : "is not"
);
}else
{
DPRINTFI("Cluster hbs last updated %d secs ago, storage-0 is not provisioned, "
"from controller-0: %d nodes enabled, %d nodes reachable, "
"from controller-1: %d nodes enabled, %d nodes reachable",
DPRINTFI("Cluster hbs last updated %d secs ago, storage-0 is not provisioned,\n"
"from controller-0: SM %s, %d nodes enabled, %d nodes reachable,\n"
"from controller-1: SM %s, %d nodes enabled, %d nodes reachable",
secs_since_update,
state.controllers[0].sm_heartbeat_fail ? "FAILED": "ok ",
state.controllers[0].number_of_node_enabled,
state.controllers[0].number_of_node_reachable,
state.controllers[1].sm_heartbeat_fail ? "FAILED": "ok ",
state.controllers[1].number_of_node_enabled,
state.controllers[1].number_of_node_reachable
);
@ -117,6 +123,10 @@ int SmClusterHbsInfoMsg::_sock = -1;
SmClusterHbsStateT SmClusterHbsInfoMsg::_cluster_hbs_state_current;
SmClusterHbsStateT SmClusterHbsInfoMsg::_cluster_hbs_state_previous;
SmClusterHbsInfoMsg::hbs_query_respond_callback SmClusterHbsInfoMsg::_callbacks;
int SmClusterHbsInfoMsg::this_controller_index = -1;
int SmClusterHbsInfoMsg::peer_controller_index = -1;
char SmClusterHbsInfoMsg::server_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1] = {0};
char SmClusterHbsInfoMsg::client_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1] = {0};
const SmClusterHbsStateT& SmClusterHbsInfoMsg::get_current_state()
{
@ -128,6 +138,47 @@ const SmClusterHbsStateT& SmClusterHbsInfoMsg::get_previous_state()
return _cluster_hbs_state_previous;
}
int SmClusterHbsInfoMsg::get_peer_controller_index()
{
if(peer_controller_index == -1)
{
get_controller_index();
}
return peer_controller_index;
}
int SmClusterHbsInfoMsg::get_this_controller_index()
{
if(this_controller_index == -1)
{
get_controller_index();
}
return this_controller_index;
}
SmErrorT SmClusterHbsInfoMsg::get_controller_index()
{
char host_name[SM_NODE_NAME_MAX_CHAR];
SmErrorT error = sm_node_utils_get_hostname(host_name);
if( SM_OKAY != error )
{
DPRINTFE( "Failed to get hostname, error=%s.",
sm_error_str( error ) );
return SM_FAILED;
}
if(0 == strncmp(SM_NODE_CONTROLLER_0_NAME, host_name, sizeof(SM_NODE_CONTROLLER_0_NAME)))
{
this_controller_index = 0;
peer_controller_index = 1;
}else
{
this_controller_index = 1;
peer_controller_index = 0;
}
return SM_OKAY;
}
bool SmClusterHbsInfoMsg::_process_cluster_hbs_history(mtce_hbs_cluster_history_type history, SmClusterHbsStateT& state)
{
if(history.controller >= max_controllers)
@ -156,6 +207,13 @@ bool SmClusterHbsInfoMsg::_process_cluster_hbs_history(mtce_hbs_cluster_history_
SmClusterHbsInfoT& controller_state = state.controllers[history.controller];
controller_state.storage0_responding = history.storage0_responding;
controller_state.sm_heartbeat_fail = (history.sm_heartbeat_fail == 1);
if(controller_state.sm_heartbeat_fail)
{
const char* controllers[] = {"controller-0", "controller-1"};
DPRINTFI("%s SM to hbsAgent alive pulse failed.", controllers[history.controller]);
}
if(entry.hosts_responding > controller_state.number_of_node_reachable)
{
controller_state.number_of_node_reachable = entry.hosts_responding;
@ -241,7 +299,7 @@ void SmClusterHbsInfoMsg::_cluster_hbs_info_msg_received( int selobj, int64_t us
}
}
SmErrorT SmClusterHbsInfoMsg::_get_address(const char* port_key, struct sockaddr_in* addr)
SmErrorT SmClusterHbsInfoMsg::_get_address(struct sockaddr_in* addr)
{
struct addrinfo *address = NULL;
struct addrinfo hints;
@ -254,17 +312,10 @@ SmErrorT SmClusterHbsInfoMsg::_get_address(const char* port_key, struct sockaddr
hints.ai_addr = NULL;
hints.ai_next = NULL;
char port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
if( SM_OKAY != sm_configuration_table_get(port_key, port, sizeof(port) - 1) )
{
DPRINTFE("Runtime error: system configuration %s undefined", port_key);
return SM_FAILED;
}
int result = getaddrinfo(LOOPBACK_IP, port, &hints, &address);
int result = getaddrinfo(LOOPBACK_IP, server_port, &hints, &address);
if(result != 0)
{
DPRINTFE("Failed to get addrinfo %s:%s", LOOPBACK_IP, port);
DPRINTFE("Failed to get addrinfo %s:%s", LOOPBACK_IP, server_port);
return SM_FAILED;
}
@ -285,15 +336,8 @@ static SmSimpleAction _query_hbs_cluster_info_action("send hbs-cluster query", s
// trigger a query of cluster hbs info.
// return true if request sent successfully, false otherwise.
// ========================
bool SmClusterHbsInfoMsg::cluster_hbs_info_query(cluster_hbs_query_ready_callback callback)
bool SmClusterHbsInfoMsg::cluster_hbs_info_query(cluster_hbs_query_ready_callback callback, bool alive_pulse)
{
char server_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
if( SM_OKAY != sm_configuration_table_get(SM_SERVER_PORT_KEY, server_port, sizeof(server_port) - 1) )
{
DPRINTFE("Runtime error: system configuration %s undefined", SM_SERVER_PORT_KEY);
return false;
}
int port = atoi(server_port);
if(0 > port)
{
@ -306,18 +350,24 @@ bool SmClusterHbsInfoMsg::cluster_hbs_info_query(cluster_hbs_query_ready_callbac
struct timespec ts;
{
mutex_holder holder(&_mutex);
if(0 != clock_gettime(CLOCK_REALTIME, &ts))
if(alive_pulse)
{
DPRINTFE("Failed to get realtime");
reqid = (unsigned short)1;
reqid = 0;
}else
{
unsigned short* v = (unsigned short*)(&ts.tv_nsec);
reqid = (*v) % 0xFFFE + 1;
if(0 != clock_gettime(CLOCK_REALTIME, &ts))
{
DPRINTFE("Failed to get realtime");
reqid = (unsigned short)1;
}else
{
unsigned short* v = (unsigned short*)(&ts.tv_nsec);
reqid = (*v) % 0xFFFE + 1;
}
}
struct sockaddr_in addr;
if(SM_OKAY != _get_address(SM_SERVER_PORT_KEY, &addr))
if(SM_OKAY != _get_address(&addr))
{
DPRINTFE("Failed to get address");
return false;
@ -325,7 +375,10 @@ bool SmClusterHbsInfoMsg::cluster_hbs_info_query(cluster_hbs_query_ready_callbac
int msg_size = snprintf(query, sizeof(query), json_fmt, reqid);
DPRINTFD("send %d bytes %s", msg_size, query);
if (reqid != 0)
{
DPRINTFI("send hbs cluster query [%d]", reqid);
}
if(0 > sendto(_sock, query, msg_size, 0, (sockaddr*)&addr, sizeof(addr)))
{
DPRINTFE("Failed to send msg. Error %s", strerror(errno));
@ -339,6 +392,11 @@ bool SmClusterHbsInfoMsg::cluster_hbs_info_query(cluster_hbs_query_ready_callbac
return true;
}
bool SmClusterHbsInfoMsg::send_alive_pulse()
{
return cluster_hbs_info_query(NULL, true);
}
SmErrorT SmClusterHbsInfoMsg::open_socket()
{
struct addrinfo *address = NULL;
@ -353,28 +411,13 @@ SmErrorT SmClusterHbsInfoMsg::open_socket()
hints.ai_next = NULL;
struct sockaddr_in addr;
char client_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
char server_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
if( SM_OKAY != sm_configuration_table_get(SM_CLIENT_PORT_KEY, client_port, sizeof(client_port) - 1) )
int port = atoi(server_port);
if(0 > port)
{
DPRINTFE("Runtime error: system configuration %s undefined", SM_CLIENT_PORT_KEY);
DPRINTFE("Invalid configuration %s: %s", SM_SERVER_PORT_KEY, server_port);
return SM_FAILED;
}
if( SM_OKAY != sm_configuration_table_get(SM_SERVER_PORT_KEY, server_port, sizeof(server_port) - 1) )
{
DPRINTFE("Runtime error: system configuration %s undefined", SM_SERVER_PORT_KEY);
return SM_FAILED;
}else
{
int port = atoi(server_port);
if(0 > port)
{
DPRINTFE("Invalid configuration %s: %s", SM_SERVER_PORT_KEY, server_port);
return SM_FAILED;
}
}
int result = getaddrinfo(LOOPBACK_IP, client_port, &hints, &address);
if(result != 0)
{
@ -441,6 +484,39 @@ SmErrorT SmClusterHbsInfoMsg::initialize()
return SM_FAILED;
}
error = get_controller_index();
if(SM_OKAY != error)
{
DPRINTFE("Failed to get controller index");
return SM_FAILED;
}
if( SM_OKAY != sm_configuration_table_get(SM_SERVER_PORT_KEY, server_port, sizeof(server_port) - 1) )
{
DPRINTFE("Runtime error: system configuration %s undefined", SM_SERVER_PORT_KEY);
return SM_FAILED;
}
int port = atoi(server_port);
if(0 > port)
{
DPRINTFE("Runtime error: Invalid configuration %s: %s", SM_SERVER_PORT_KEY, server_port);
return SM_FAILED;
}
if( SM_OKAY != sm_configuration_table_get(SM_CLIENT_PORT_KEY, client_port, sizeof(client_port) - 1) )
{
DPRINTFE("Runtime error: system configuration %s undefined", SM_CLIENT_PORT_KEY);
return SM_FAILED;
}
port = atoi(client_port);
if(0 > port)
{
DPRINTFE("Runtime error: Invalid configuration %s: %s", SM_CLIENT_PORT_KEY, client_port);
return SM_FAILED;
}
error = open_socket();
if(SM_OKAY != error)
{

18
service-mgmt/sm/src/sm_cluster_hbs_info_msg.h
View File

@ -12,6 +12,7 @@
#include "sm_types.h"
#include "sm_timer.h"
#include "sm_util_types.h"
#include "sm_limits.h"
// ****************************************************************************
// struct SmClusterHbsInfoT
@ -22,9 +23,11 @@ struct _SmClusterHbsInfoT
bool storage0_responding;
int number_of_node_reachable;
int number_of_node_enabled;
bool sm_heartbeat_fail;
_SmClusterHbsInfoT() : storage0_responding(false),
number_of_node_reachable(0),
number_of_node_enabled(0)
number_of_node_enabled(0),
sm_heartbeat_fail(false)
{
}
};
@ -74,8 +77,11 @@ class SmClusterHbsInfoMsg
static SmErrorT finalize();
static const SmClusterHbsStateT& get_current_state();
static const SmClusterHbsStateT& get_previous_state();
static bool cluster_hbs_info_query(cluster_hbs_query_ready_callback callback = NULL);
static bool cluster_hbs_info_query(cluster_hbs_query_ready_callback callback = NULL, bool alive_pulse = false);
static bool send_alive_pulse();
static void dump_hbs_record(FILE* fp);
static int get_peer_controller_index();
static int get_this_controller_index();
private:
static int _sock;
@ -85,11 +91,17 @@ class SmClusterHbsInfoMsg
static SmClusterHbsStateT _cluster_hbs_state_previous;
static hbs_query_respond_callback _callbacks;
static SmErrorT open_socket();
static SmErrorT get_controller_index();
static SmErrorT _get_address(const char* port_key, struct sockaddr_in* addr);
static SmErrorT _get_address(struct sockaddr_in* addr);
static void _cluster_hbs_info_msg_received( int selobj, int64_t user_data );
static bool _process_cluster_hbs_history(mtce_hbs_cluster_history_type history,
SmClusterHbsStateT& state);
static char server_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
static char client_port[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
static int peer_controller_index;
static int this_controller_index;
};
#endif // __SM_CLUSTER_HBS_INFO_MSG_H__

96
service-mgmt/sm/src/sm_failover.c
View File

@ -11,6 +11,9 @@
#include <time.h>
#include <stdio.h>
#include <json-c/json.h>
#include <errno.h>
#include <unistd.h>
#include <fcntl.h>
#include "sm_service_domain_interface_table.h"
#include "sm_debug.h"
@ -44,6 +47,8 @@
#define SM_FAILOVER_RECOVERY_INTERVAL_IN_SEC 100
#define SM_FAILOVER_INTERFACE_STATE_REPORT_INTERVAL_MS 20000
const char* RESET_PEER_NOW = "/var/run/.sm_reset_peer";
typedef enum
{
SM_FAILOVER_ACTION_RESULT_OK,
@ -816,12 +821,25 @@ static SmErrorT sm_ensure_leader_scheduler()
}
return error;
}
bool file_exist(const char* filename)
{
if( 0 != access(filename, F_OK ) )
{
if( ENOENT == errno )
{
return false;
}
}
return true;
}
// ****************************************************************************
// Failover - set system to scheduled status
// ==================
SmErrorT sm_failover_set_system(const SmSystemFailoverStatus& failover_status)
{
//valid combinations of target system scheduling are:
// valid combinations of target system scheduling are:
// active/standby
// active/failed
SmFailoverActionResultT result;
@ -829,6 +847,47 @@ SmErrorT sm_failover_set_system(const SmSystemFailoverStatus& failover_status)
host_target_state = failover_status.get_host_schedule_state();
peer_target_state = failover_status.get_peer_schedule_state();
SmHeartbeatStateT heartbeat_state = failover_status.get_heartbeat_state();
if(SM_NODE_STATE_ACTIVE == host_target_state &&
SM_NODE_STATE_FAILED == peer_target_state &&
failover_status.peer_stall())
{
int seconds_to_wait = sm_failover_get_reset_peer_wait_time();
int fd = open(RESET_PEER_NOW, O_RDWR | O_CREAT, S_IRUSR | S_IRGRP | S_IROTH);
if( 0 > fd )
{
DPRINTFE("Failed to create file (%s), error=%s.", RESET_PEER_NOW, strerror(errno) );
}else
{
close(fd);
}
for(int i = 0; i < seconds_to_wait * 10; i ++)
{
//wait up to 30 seconds for mtce to reset peer.
if(!file_exist(RESET_PEER_NOW))
{
DPRINTFI("%s is gone.", RESET_PEER_NOW);
break;
}else
{
usleep(100000); // 100ms
if(i % 10 == 0)
{
// log every second
DPRINTFI("Wait for %s to be removed", RESET_PEER_NOW);
}
}
}
if(file_exist(RESET_PEER_NOW))
{
DPRINTFE("%s still exists after %d seconds. Start activating the controller.",
RESET_PEER_NOW, seconds_to_wait);
}
}
if(SM_HEARTBEAT_OK != heartbeat_state)
{
if(SM_OKAY != sm_ensure_leader_scheduler())
@ -1102,6 +1161,9 @@ SmErrorT sm_failover_disable_peer()
{
DPRINTFE("Failed to disable peer %s", _peer_name);
return SM_FAILED;
}else
{
DPRINTFI("Request mtce to reset peer");
}
return SM_OKAY;
}
@ -1138,6 +1200,8 @@ SmFailoverInterfaceStateT sm_failover_get_interface_info(SmInterfaceTypeT interf
static void sm_failover_interface_change_callback(
SmHwInterfaceChangeDataT* if_change )
{
SmFailoverInterfaceInfo* iter;
SmServiceDomainInterfaceT* interface;
switch ( if_change->interface_state )
{
case SM_INTERFACE_STATE_DISABLED:
@ -1147,8 +1211,20 @@ static void sm_failover_interface_change_callback(
sm_failover_interface_up(if_change->interface_name);
break;
default:
DPRINTFI("Interface %s state changed to %d",
if_change->interface_name, if_change->interface_state);
// skip logging the state change of interfaces that are not monitored
// by SM.
for(iter = _my_if_list; iter < _my_if_list + _total_interfaces; iter ++)
{
interface = iter->get_interface();
if(strncmp(interface->interface_name, if_change->interface_name,
sizeof(if_change->interface_name)) == 0)
{
DPRINTFI("Interface %s state changed to %d",
if_change->interface_name, if_change->interface_state);
break;
}
}
break;
}
}
@ -1183,7 +1259,7 @@ SmErrorT sm_failover_initialize( void )
_system_mode = sm_node_utils_get_system_mode();
DPRINTFI("System mode %s", sm_system_mode_str(_system_mode));
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle, true );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",
@ -1292,12 +1368,6 @@ SmErrorT sm_failover_initialize( void )
return SM_FAILED;
}
error = SmClusterHbsInfoMsg::initialize();
if(SM_OKAY != error)
{
DPRINTFE("Failed to initialize cluster hbs info messaging");
}
return SM_OKAY;
}
// ****************************************************************************
@ -1311,12 +1381,6 @@ SmErrorT sm_failover_finalize( void )
SmErrorT error;
error = SmClusterHbsInfoMsg::finalize();
if(SM_OKAY != error)
{
DPRINTFE("Failed to finalize cluster hbs info messaging");
}
sm_timer_deregister( failover_audit_timer_id );
if( NULL != _sm_db_handle )
{

2
service-mgmt/sm/src/sm_failover.h
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2017 Wind River Systems, Inc.
// Copyright (c) 2017-2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//

38
service-mgmt/sm/src/sm_failover_fail_pending_state.cpp
View File

@ -6,6 +6,7 @@
#include "sm_failover_fail_pending_state.h"
#include <stdlib.h>
#include <unistd.h>
#include "sm_configuration_table.h"
#include "sm_cluster_hbs_info_msg.h"
#include "sm_types.h"
#include "sm_limits.h"
@ -18,8 +19,26 @@
#include "sm_node_api.h"
#include "sm_worker_thread.h"
static const int FAIL_PENDING_TIMEOUT = 2000; // 2seconds
static const int DELAY_QUERY_HBS_MS = FAIL_PENDING_TIMEOUT - 200; // give 200ms for hbs agent to respond
static const int FAIL_PENDING_TIMEOUT_DEFAULT = 2000; // 2 seconds
static const int FAIL_PENDING_TIMEOUT_MIN = 1200;
static inline int get_failpending_timeout()
{
int fail_pending_timeout =0;
char buf[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
if( SM_OKAY == sm_configuration_table_get("FAILPENDING_TIMEOUT_MS", buf, sizeof(buf) - 1) )
{
fail_pending_timeout = atoi(buf);
}
if(fail_pending_timeout < FAIL_PENDING_TIMEOUT_MIN)
{
fail_pending_timeout = FAIL_PENDING_TIMEOUT_DEFAULT;
}
return fail_pending_timeout;
}
static const int fail_pending_timeout = 0;
static SmTimerIdT action_timer_id = SM_TIMER_ID_INVALID;
static const int RESET_TIMEOUT = 10 * 1000; // 10 seconds for a reset command to reboot a node
@ -187,8 +206,8 @@ SmFailoverFailPendingState::~SmFailoverFailPendingState()
SmErrorT SmFailoverFailPendingState::event_handler(SmFailoverEventT event, const ISmFSMEventData* event_data)
{
//SmFSMEventDataTypeT event_data_type = event_data->get_event_data_type();
bool duplex = false;
bool blind_guess = false;
switch (event)
{
case SM_FAILOVER_EVENT_IF_STATE_CHANGED:
@ -230,9 +249,10 @@ SmErrorT SmFailoverFailPendingState::event_handler(SmFailoverEventT event, const
if(healthy)
{
blind_guess_scenario_start();
blind_guess = true;
}
}
else
if( !blind_guess )
{
SmSystemFailoverStatus& failover_status = SmSystemFailoverStatus::get_status();
SmErrorT error = sm_failover_ss_get_survivor(failover_status);
@ -288,6 +308,10 @@ bool SmFailoverFailPendingState::_fail_pending_timeout(
SmErrorT SmFailoverFailPendingState::enter_state()
{
SmFSMState::enter_state();
DPRINTFI("Pre failure hbs cluster info:");
const SmClusterHbsStateT& cluster_hbs_state = SmClusterHbsInfoMsg::get_current_state();
log_cluster_hbs_state(cluster_hbs_state);
SmErrorT error = this->_register_timer();
if(SM_OKAY != error)
{
@ -299,6 +323,7 @@ SmErrorT SmFailoverFailPendingState::enter_state()
void _cluster_hbs_response_callback()
{
const SmClusterHbsStateT& cluster_hbs_state = SmClusterHbsInfoMsg::get_current_state();
DPRINTFI("Fail-pending timeout cluster info:");
log_cluster_hbs_state(cluster_hbs_state);
SmSystemFailoverStatus::get_status().set_cluster_hbs_state(cluster_hbs_state);
}
@ -319,13 +344,14 @@ SmErrorT SmFailoverFailPendingState::_register_timer()
this->_deregister_timer();
}
error = sm_timer_register(timer_name, FAIL_PENDING_TIMEOUT,
int fail_pending_timeout = get_failpending_timeout();
error = sm_timer_register(timer_name, fail_pending_timeout,
SmFailoverFailPendingState::_fail_pending_timeout,
0, &this->_pending_timer_id);
const char* delay_query_hbs_timer_name = "DELAY QUERY HBS";
error = sm_timer_register(delay_query_hbs_timer_name, DELAY_QUERY_HBS_MS,
error = sm_timer_register(delay_query_hbs_timer_name, fail_pending_timeout - 200,
SmFailoverFailPendingState::_delay_query_hbs_timeout,
0, &this->_delay_query_hbs_timer_id);

317
service-mgmt/sm/src/sm_failover_failed_state.cpp
View File

@ -1,26 +1,274 @@
//
// Copyright (c) 2018 Wind River Systems, Inc.
// Copyright (c) 2018-2021 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
#include "sm_failover_failed_state.h"
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <limits.h>
#include <sys/types.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/stat.h>
#include <errno.h>
#include <fcntl.h>
#include "sm_failover_failed_state.h"
#include "sm_types.h"
#include "sm_debug.h"
#include "sm_node_utils.h"
#include "sm_node_api.h"
#include "sm_failover.h"
#include "sm_failover_fsm.h"
#include "sm_failover_ss.h"
#include "sm_failover_utils.h"
#include "sm_cluster_hbs_info_msg.h"
extern bool is_cluster_host_interface_configured( void );
// Failover Failed Recovery Audit period = 5 seconds
static const int FAILED_STATE_AUDIT_PERIOD = 5000;
// Recovery log throttle threshold - 1 log every minute
static const int SM_FAILOVER_FAILED_LOG_THROTTLE_THLD = 12;
// processes to restart over a failover failed recovery
#define MAX_RESTART_PROCESS_NAME_LEN 10
#define PROCESS_SM ((const char *)("sm"))
static struct timespec start_time;
// Failover Failed state class constructor
SmFailoverFailedState::SmFailoverFailedState(SmFailoverFSM& fsm) : SmFSMState(fsm)
{
this->_failed_state_audit_timer_id = SM_TIMER_ID_INVALID;
}
// The 'Failover Failed' state destructor
// - stops the recovery audit if needed
SmFailoverFailedState::~SmFailoverFailedState()
{
this->_deregister_timer();
}
// Failover Failed state entry class member function
// - starts the Failover Failed state recovery audit timer
SmErrorT SmFailoverFailedState::enter_state()
{
SmFSMState::enter_state();
DPRINTFE("********************************************************");
DPRINTFE("Entering Failover Failed state ; recovery audit started ");
DPRINTFE("********************************************************");
clock_gettime(CLOCK_MONOTONIC_RAW, &start_time);
DPRINTFI("Wait for %d seconds to start failed recovery audit",
sm_failover_get_reset_peer_wait_time());
SmErrorT error = this->_register_timer();
if(SM_OKAY != error)
{
DPRINTFE("Failed to register failed state timer. Error %s", sm_error_str(error));
}
return error;
}
// Failover Failed state audit timer handler
bool SmFailoverFailedState::_failed_state_audit(
SmTimerIdT timer_id, int64_t user_data)
{
// wait enough time for peer to be reset to avoid conflict decision, i.e being reset
struct timespec now;
clock_gettime(CLOCK_REALTIME, &now);
if(now.tv_sec - start_time.tv_sec - 1 > sm_failover_get_reset_peer_wait_time())
{
SmFailoverFSM::get_fsm().send_event(SM_FAILOVER_EVENT_FAILED_RECOVERY_AUDIT, NULL);
}
return true ;
}
// Issue a self restart through pmon-restart service
static bool sm_failover_failed_process_restart( const char * process )
{
DPRINTFI( "Issuing controlled process restart ; pmon-restart %s", process);
pid_t pid = fork();
if( 0 > pid )
{
DPRINTFE( "Failed to fork 'pmond-restart %s' request, error=%s.",
process, strerror( errno ) );
return( true );
}
else if( 0 == pid )
{
// set the arguement array for execv
char pmon_restart_cmd[] = "/usr/local/sbin/pmon-restart";
char pmon_restart_process[MAX_RESTART_PROCESS_NAME_LEN] ;
snprintf(&pmon_restart_process[0], MAX_RESTART_PROCESS_NAME_LEN, "%s", process);
char* pmon_restart_argv[3] ;
pmon_restart_argv[0] = pmon_restart_cmd;
pmon_restart_argv[1] = pmon_restart_process;
pmon_restart_argv[2] = NULL;
// Add the path to socat for pmon-restart
char path[] = "PATH=/usr/bin:$PATH";
char* pmon_restart_env[2] ;
pmon_restart_env[0] = path;
pmon_restart_env[1] = NULL;
setpgid( 0, 0 );
struct rlimit file_limits;
if( 0 == getrlimit( RLIMIT_NOFILE, &file_limits ) )
{
unsigned int fd_i;
for( fd_i=0; fd_i < file_limits.rlim_cur; ++fd_i )
{
close( fd_i );
}
open( "/dev/null", O_RDONLY ); // stdin
open( "/dev/null", O_WRONLY ); // stdout
open( "/dev/null", O_WRONLY ); // stderr
}
execve( pmon_restart_argv[0], pmon_restart_argv, pmon_restart_env );
// Shouldn't get this far, else there was an error.
exit(-1);
}
return( false );
}
// Failover Failed recovery criteria checker
static bool sm_failover_failed_recovery_criteria_met( void )
{
bool criteria_met = false ;
SmFailoverInterfaceStateT oam_state, mgmt_state, cluster_host_state;
oam_state = sm_failover_get_interface_info(SM_INTERFACE_OAM);
mgmt_state = sm_failover_get_interface_info(SM_INTERFACE_MGMT);
const SmClusterHbsStateT& cluster_hbs_state = SmClusterHbsInfoMsg::get_current_state();
int peer_controller_index = SmClusterHbsInfoMsg::get_peer_controller_index();
// If peer has stalled, do not recover until peer recovers from stall
if(!cluster_hbs_state.controllers[peer_controller_index].sm_heartbeat_fail)
{
if ( is_cluster_host_interface_configured() )
{
cluster_host_state = sm_failover_get_interface_info(SM_INTERFACE_CLUSTER_HOST);
if ((( oam_state == SM_FAILOVER_INTERFACE_OK ) || ( oam_state == SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT )) &&
(( mgmt_state == SM_FAILOVER_INTERFACE_OK ) || ( mgmt_state == SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT )) &&
(( cluster_host_state == SM_FAILOVER_INTERFACE_OK ) || ( cluster_host_state == SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT )))
{
criteria_met = true ;
}
}
else if ((( oam_state == SM_FAILOVER_INTERFACE_OK ) || ( oam_state == SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT )) &&
(( mgmt_state == SM_FAILOVER_INTERFACE_OK ) || ( mgmt_state == SM_FAILOVER_INTERFACE_MISSING_HEARTBEAT )))
{
criteria_met = true ;
}
}
DPRINTFI("Oam:%s ; Mgmt:%s ; Cluster:%s ; recovery criteria met: %s",
sm_failover_interface_state_str(oam_state),
sm_failover_interface_state_str(mgmt_state),
sm_failover_interface_state_str(cluster_host_state),
criteria_met ? "Yes" : "No");
return (criteria_met);
}
SmErrorT proceed_recovery()
{
SmErrorT error;
char peer_name[SM_NODE_NAME_MAX_CHAR];
char host_name[SM_NODE_NAME_MAX_CHAR];
// delete peer node
error = sm_node_api_get_peername(peer_name);
if(SM_OKAY != error)
{
DPRINTFI("Cannot retrieve peer's hostname, error %s", sm_error_str(error));
return error;
}
error = sm_node_api_delete_node(peer_name);
if(SM_OKAY != error)
{
DPRINTFI("Failed to delete peer %s, error %s", peer_name, sm_error_str(error));
return error;
}else
{
DPRINTFI("Peer %s is deleted.", peer_name);
}
// enable host
error = sm_node_api_get_hostname(host_name);
if(SM_OKAY != error)
{
DPRINTFI("Cannot retrieve hostname, error %s", sm_error_str(error));
return error;
}
error = sm_node_api_recover_node(host_name);
if(SM_OKAY != error)
{
DPRINTFI("Failed to recover %s, error %s", host_name, sm_error_str(error));
return error;
}else
{
DPRINTFI("Host %s is recovered.", host_name);
}
sm_node_utils_reset_unhealthy_flag();
DPRINTFI("Unhealthy flag is removed");
return SM_OKAY;
}
// The 'Failover Failed' state recovery audit handler
SmErrorT SmFailoverFailedState::event_handler(SmFailoverEventT event, const ISmFSMEventData* event_data)
{
// Currently the only supported scenario to recover from failure is
// reboot triggered by mtce.
// So once entering failed state, wait for reboot to reenter the normal state.
SmErrorT error;
event_data=event_data;
switch (event)
{
case SM_FAILOVER_EVENT_IF_STATE_CHANGED:
// event will be fired, but couldn't bring fsm state back to normal
case SM_FAILOVER_EVENT_FAILED_RECOVERY_AUDIT:
{
if ( sm_failover_failed_recovery_criteria_met() )
{
DPRINTFI("************************************");
DPRINTFI("** Failover Failed state recovery **");
DPRINTFI("************************************");
error = proceed_recovery();
if(SM_OKAY != error)
{
DPRINTFE("Cannot recover from failed state");
}else
{
sm_failover_failed_process_restart(PROCESS_SM);
for ( int i = 0 ; i < 10 ; i++ )
{
// waiting for shutdown
sleep(1);
}
DPRINTFE("Restart did not occur ; reinstating unhealthy flag ; recovery will retry");
sm_node_utils_set_unhealthy();
}
}
else if ( ++_log_throttle > 1 )
{
if ( _log_throttle > SM_FAILOVER_FAILED_LOG_THROTTLE_THLD )
_log_throttle = 0 ;
}
else
{
DPRINTFI("Failover Failed state recovery monitor");
}
break;
}
default:
DPRINTFE("Runtime error, unexpected event %s, at state %s",
sm_failover_event_str(event),
@ -28,3 +276,58 @@ SmErrorT SmFailoverFailedState::event_handler(SmFailoverEventT event, const ISmF
}
return SM_OKAY;
}
// Start the 'Failover Failed' state recovery audit
SmErrorT SmFailoverFailedState::_register_timer()
{
SmErrorT error;
const char* timer_name = "FAILED STATE AUDIT TIMER";
if(SM_TIMER_ID_INVALID != this->_failed_state_audit_timer_id)
this->_deregister_timer();
error = sm_timer_register(timer_name, FAILED_STATE_AUDIT_PERIOD,
SmFailoverFailedState::_failed_state_audit,
0, &this->_failed_state_audit_timer_id);
return error;
}
// Stop the 'Failover Failed' state recovery audit
SmErrorT SmFailoverFailedState::_deregister_timer()
{
SmErrorT error = SM_OKAY;
if(SM_TIMER_ID_INVALID != this->_failed_state_audit_timer_id)
{
error = sm_timer_deregister(this->_failed_state_audit_timer_id);
if( SM_OKAY != error )
{
DPRINTFE( "Failed to cancel failed timer, error=%s.",
sm_error_str( error ) );
}else
{
this->_failed_state_audit_timer_id = SM_TIMER_ID_INVALID;
}
}
return error;
}
SmErrorT SmFailoverFailedState::exit_state()
{
SmErrorT error = this->_deregister_timer();
if(SM_OKAY != error)
{
DPRINTFE("Failed to deregister fail failed timer. Error %s", sm_error_str(error));
}
if(SM_TIMER_ID_INVALID != _failed_state_audit_timer_id)
{
error = sm_timer_deregister(_failed_state_audit_timer_id);
_failed_state_audit_timer_id = SM_TIMER_ID_INVALID;
if( SM_OKAY != error)
{
DPRINTFE("Failed to deregister action timer. Error %s", sm_error_str(error));
}
}
SmFSMState::exit_state();
return error;
}

16
service-mgmt/sm/src/sm_failover_failed_state.h
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2018 Wind River Systems, Inc.
// Copyright (c) 2020 Wind River Systems, Inc.
//
// SPDX-License-Identifier: Apache-2.0
//
@ -12,12 +12,22 @@
class SmFailoverFailedState : public SmFSMState
{
public:
SmFailoverFailedState(SmFailoverFSM& fsm) : SmFSMState(fsm){}
SmFailoverFailedState(SmFailoverFSM& fsm);
virtual ~SmFailoverFailedState();
SmErrorT enter_state();
SmErrorT exit_state();
protected:
SmErrorT event_handler(SmFailoverEventT event, const ISmFSMEventData* event_data);
private:
SmTimerIdT _failed_state_audit_timer_id;
static bool _failed_state_audit(SmTimerIdT timer_id, int64_t user_data);
SmErrorT _register_timer();
SmErrorT _deregister_timer();
int _log_throttle ;
};
#endif //__SM_FAILOVER_FAILED_STATE_H__
#endif //__SM_FAILOVER_FAILED_STATE_H__

35
service-mgmt/sm/src/sm_failover_ss.c
View File

@ -384,32 +384,25 @@ SmErrorT sm_failover_ss_get_survivor(const SmSystemStatusT& system_status, SmSys
has_cluser_info = false;
}
DPRINTFI("cluster info (%s), maximum available nodes %d", has_cluser_info ? "yes": "no", max_nodes_available);
if(has_cluser_info && max_nodes_available > 1)
{
DPRINTFD("storage-0 is %s", expect_storage_0 ? "enabled":"not enabled");
int this_controller_index, peer_controller_index;
char host_name[SM_NODE_NAME_MAX_CHAR];
SmErrorT error = sm_node_utils_get_hostname(host_name);
if( SM_OKAY != error )
{
DPRINTFE( "Failed to get hostname, error=%s.",
sm_error_str( error ) );
return SM_FAILED;
}
if(0 == strncmp(SM_NODE_CONTROLLER_0_NAME, host_name, sizeof(SM_NODE_CONTROLLER_0_NAME)))
{
this_controller_index = 0;
peer_controller_index = 1;
}else
{
this_controller_index = 1;
peer_controller_index = 0;
}
int this_controller_index = SmClusterHbsInfoMsg::get_this_controller_index();
int peer_controller_index = SmClusterHbsInfoMsg::get_peer_controller_index();
bool survivor_selected = false;
if(expect_storage_0)
selection.set_peer_stall(false);
if(current_cluster_hbs_state.controllers[peer_controller_index].sm_heartbeat_fail)
{
// peer sm not sending alive pulse, failed
DPRINTFI("Peer controller has stalled.");
selection.set_host_schedule_state(SM_NODE_STATE_ACTIVE);
selection.set_peer_schedule_state(SM_NODE_STATE_FAILED);
survivor_selected = true;
selection.set_peer_stall(true);
}
else if(expect_storage_0)
{
if(current_cluster_hbs_state.controllers[this_controller_index].storage0_responding &&
!current_cluster_hbs_state.controllers[peer_controller_index].storage0_responding)

9
service-mgmt/sm/src/sm_failover_ss.h
View File

@ -69,6 +69,14 @@ class SmSystemFailoverStatus
inline SmHeartbeatStateT get_heartbeat_state() const {
return _heartbeat_state;
}
inline bool peer_stall() const {
return _peer_stall;
}
inline void set_peer_stall(bool stall) {
_peer_stall = stall;
}
void set_host_schedule_state(SmNodeScheduleStateT state);
void set_host_pre_failure_schedule_state(SmNodeScheduleStateT state);
void set_cluster_hbs_state(const SmClusterHbsStateT& state);
@ -93,6 +101,7 @@ class SmSystemFailoverStatus
SmNodeScheduleStateT _peer_schedule_state;
SmClusterHbsStateT _cluster_hbs_state;
SmHeartbeatStateT _heartbeat_state;
bool _peer_stall;
static const char filename[];
static const char file_format[];

25
service-mgmt/sm/src/sm_failover_utils.c
View File

@ -8,6 +8,7 @@
#include "sm_service_domain_member_table.h"
#include "sm_service_domain_assignment_table.h"
#include "sm_service_domain_table.h"
#include "sm_configuration_table.h"
#define SM_NODE_STAY_FAILED_FILE "/var/run/.sm_stay_fail"
@ -139,3 +140,27 @@ bool sm_failover_utils_is_stayfailed()
return false;
}
// ****************************************************************************
// ****************************************************************************
// Failover Utilities - get wait time for peer to be reset
// ==============================
int sm_failover_get_reset_peer_wait_time()
{
char buf[SM_CONFIGURATION_VALUE_MAX_CHAR + 1];
const int RESET_PEER_WAIT_TIME_MIN_SEC = 1;
const int RESET_PEER_WAIT_TIME_DEFAULT_SEC = 30;
int seconds_to_wait = 0;
if( SM_OKAY == sm_configuration_table_get("RESET_PEER_WAIT_TIMEOUT_SEC", buf, sizeof(buf) - 1) )
{
seconds_to_wait = atoi(buf);
}
if( seconds_to_wait < RESET_PEER_WAIT_TIME_MIN_SEC)
{
seconds_to_wait = RESET_PEER_WAIT_TIME_DEFAULT_SEC;
}
DPRINTFI("Peer reset wait time %d sec", seconds_to_wait);
return seconds_to_wait;
}
// ****************************************************************************

4
service-mgmt/sm/src/sm_failover_utils.h
View File

@ -36,6 +36,10 @@ extern SmErrorT sm_failover_utils_reset_stayfailed_flag();
extern bool sm_failover_utils_is_stayfailed();
// ****************************************************************************
// ****************************************************************************
// Failover Utilities - get wait time for peer to be reset
// ==============================
extern int sm_failover_get_reset_peer_wait_time();
#ifdef __cplusplus
}
#endif

34
service-mgmt/sm/src/sm_heartbeat_thread.c
View File

@ -35,6 +35,7 @@
#include "sm_alarm.h"
#include "sm_log.h"
#include "sm_failover.h"
#include "sm_cluster_hbs_info_msg.h"
#define SM_HEARTBEAT_THREAD_NAME "sm_heartbeat"
#define SM_HEARTBEAT_THREAD_TICK_INTERVAL_IN_MS 100
@ -106,6 +107,10 @@ static SmTimerIdT _alive_timer_id = SM_TIMER_ID_INVALID;
static char _node_name[SM_NODE_NAME_MAX_CHAR];
static SmHeartbeatMsgCallbacksT _callbacks;
static int alive_pulse_count = 0;
static int alive_pulse_fail_count = 0;
static SmTimeT alive_pulse_since = {0};
// ****************************************************************************
// Heartbeat Thread - SmNetworkAddressT ==
// =========================
@ -921,6 +926,35 @@ static bool sm_heartbeat_alive_timer( SmTimerIdT timer_id, int64_t user_data )
DPRINTFD( "Sent alive message for node (%s).", _node_name );
}
if(!SmClusterHbsInfoMsg::send_alive_pulse())
{
alive_pulse_fail_count ++;
DPRINTFE("Failed sending alive pulse to hbsAgent");
}else
{
alive_pulse_count ++;
}
if(alive_pulse_since.tv_sec == 0)
{
clock_gettime(CLOCK_MONOTONIC_RAW, &alive_pulse_since);
}else
{
SmTimeT now;
clock_gettime(CLOCK_MONOTONIC_RAW, &now);
int delta;
delta = (int)(now.tv_sec - alive_pulse_since.tv_sec);
if(delta > 60)
{
DPRINTFI("Record %d alive pulses (include %d failed) to hbsAgent since %d seconds ago",
alive_pulse_count + alive_pulse_fail_count, alive_pulse_fail_count, delta);
alive_pulse_count = 0;
alive_pulse_fail_count = 0;
alive_pulse_since = now;
}
}
DONE:
if( 0 != pthread_mutex_unlock( &_mutex ) )
{

23
service-mgmt/sm/src/sm_main_event_handler.c
View File

@ -32,7 +32,6 @@
#define SM_NODE_AUDIT_TIMER_IN_MS 1000
#define SM_INTERFACE_AUDIT_TIMER_IN_MS 1000
static SmDbHandleT* _sm_db_handle = NULL;
static SmApiCallbacksT _api_callbacks = {0};
static SmNotifyApiCallbacksT _notify_api_callbacks = {0};
static SmTimerIdT _node_audit_timer_id = SM_TIMER_ID_INVALID;
@ -322,14 +321,6 @@ SmErrorT sm_main_event_handler_initialize( void )
memset( &_api_callbacks, 0, sizeof(_api_callbacks) );
memset( &_notify_api_callbacks, 0, sizeof(_notify_api_callbacks) );
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
return( error );
}
error = sm_timer_register( "node audit",
SM_NODE_AUDIT_TIMER_IN_MS,
sm_main_event_handler_audit_node,
@ -372,7 +363,7 @@ SmErrorT sm_main_event_handler_initialize( void )
{
DPRINTFE( "Failed to release service groups, error=%s.",
sm_error_str( error ) );
return( error );
return error;
}
error = sm_api_initialize();
@ -498,18 +489,6 @@ SmErrorT sm_main_event_handler_finalize( void )
_interface_audit_timer_id = SM_TIMER_ID_INVALID;
}
if( NULL != _sm_db_handle )
{
error = sm_db_disconnect( _sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to disconnect from database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
}
_sm_db_handle = NULL;
}
return( SM_OKAY );
}
// ****************************************************************************

6
service-mgmt/sm/src/sm_msg.c
View File

@ -803,6 +803,12 @@ static int sm_send_ipv6_msg(SmServiceDomainInterfaceT* interface, SmMsgT* msg )
ipv6_dst = &(interface->network_peer_address.u.ipv6);
}
dst_addr6.sin6_addr = ipv6_dst->sin6;
if (memcmp(&in6addr_any, &dst_addr6.sin6_addr, sizeof(dst_addr6.sin6_addr)) == 0 )
{
// don't send to ::
return _MSG_NOT_SENT_TO_TARGET;
}
result = sm_msg_sendmsg_src_ipv6( interface->unicast_socket, msg, sizeof(SmMsgT),
0, &dst_addr6, &interface->network_address.u.ipv6.sin6 );

38
service-mgmt/sm/src/sm_node_api.cpp
View File

@ -742,6 +742,44 @@ SmErrorT sm_node_api_fail_node( char node_name[] )
}
// ****************************************************************************
// ****************************************************************************
// Node API - Recover Node
// ======================
SmErrorT sm_node_api_recover_node( char node_name[] )
{
SmDbNodeT node;
SmErrorT error;
error = sm_db_nodes_read( _sm_db_handle, node_name, &node );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to read node (%s) information, error=%s.",
node_name, sm_error_str( error ) );
return( error );
}
if( node.oper_state != SM_NODE_OPERATIONAL_STATE_DISABLED ||
node.avail_status != SM_NODE_AVAIL_STATUS_FAILED )
{
DPRINTFD("Not in failure mode %s", node_name);
}
DPRINTFE("Node %s is to recover from failure mode.", node_name);
error = sm_node_api_update_node(
node_name,
node.admin_state,
SM_NODE_OPERATIONAL_STATE_ENABLED,
SM_NODE_AVAIL_STATUS_AVAILABLE);
if( SM_OKAY != error )
{
DPRINTFE( "Failed to set node (%s) failed, error=%s.",
node_name, sm_error_str( error ) );
}
return( error );
}
// ****************************************************************************
// ****************************************************************************
// Node API - Delete Node
// ======================

6
service-mgmt/sm/src/sm_node_api.h
View File

@ -54,6 +54,12 @@ extern SmErrorT sm_node_api_update_node( char node_name[],
SmErrorT sm_node_api_fail_node( char node_name[] );
// ****************************************************************************
// ****************************************************************************
// Node API - Fail Node
// ======================
SmErrorT sm_node_api_recover_node( char node_name[] );
// ****************************************************************************
// ****************************************************************************
// Node API - Delete Node
// ======================

13
service-mgmt/sm/src/sm_process.c
View File

@ -54,6 +54,7 @@
#include "sm_task_affining_thread.h"
#include "sm_worker_thread.h"
#include "sm_configuration_table.h"
#include "sm_cluster_hbs_info_msg.h"
#define SM_PROCESS_DB_CHECKPOINT_INTERVAL_IN_MS 30000
#define SM_PROCESS_TICK_INTERVAL_IN_MS 200
@ -233,6 +234,12 @@ static SmErrorT sm_process_initialize( void )
return( SM_FAILED );
}
error = SmClusterHbsInfoMsg::initialize();
if(SM_OKAY != error)
{
DPRINTFE("Failed to initialize cluster hbs info messaging");
}
if (_is_aio_simplex)
{
sm_heartbeat_thread_disable_heartbeat();
@ -518,6 +525,12 @@ static SmErrorT sm_process_finalize( void )
sm_error_str( error ) );
}
error = SmClusterHbsInfoMsg::finalize();
if(SM_OKAY != error)
{
DPRINTFE("Failed to finalize cluster hbs info messaging");
}
error = sm_log_finalize();
if( SM_OKAY != error )
{

23
service-mgmt/sm/src/sm_service_action_result_table.c
View File

@ -18,7 +18,6 @@
#include "sm_db_service_action_results.h"
static SmListT* _service_action_results = NULL;
static SmDbHandleT* _sm_db_handle = NULL;
// ****************************************************************************
// Service Action Result Table - Read
@ -149,14 +148,6 @@ SmErrorT sm_service_action_result_table_initialize( void )
_service_action_results = NULL;
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
return( error );
}
error = sm_service_action_result_table_load();
if( SM_OKAY != error )
{
@ -174,22 +165,8 @@ SmErrorT sm_service_action_result_table_initialize( void )
// ======================================
SmErrorT sm_service_action_result_table_finalize( void )
{
SmErrorT error;
SM_LIST_CLEANUP_ALL( _service_action_results );
if( NULL != _sm_db_handle )
{
error = sm_db_disconnect( _sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to disconnect from database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
}
_sm_db_handle = NULL;
}
return( SM_OKAY );
}
// ****************************************************************************

2
service-mgmt/sm/src/sm_service_action_table.c
View File

@ -204,7 +204,7 @@ SmErrorT sm_service_action_table_initialize( void )
_service_actions = NULL;
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle, true );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",

2
service-mgmt/sm/src/sm_service_dependency_table.c
View File

@ -220,7 +220,7 @@ SmErrorT sm_service_dependency_table_initialize( void )
{
SmErrorT error;
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle, true );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",

2
service-mgmt/sm/src/sm_service_domain_api.c
View File

@ -149,7 +149,7 @@ SmErrorT sm_service_domain_api_initialize( void )
{
SmErrorT error;
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle, true );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",

17
service-mgmt/sm/src/sm_service_domain_interface_unknown_state.c
View File

@ -63,17 +63,20 @@ SmErrorT sm_service_domain_interface_unknown_state_event_handler(
SmServiceDomainInterfaceT* interface, SmServiceDomainInterfaceEventT event,
void* event_data[] )
{
bool enabled;
bool enabled = false;
SmErrorT error;
char reason_text[SM_LOG_REASON_TEXT_MAX_CHAR] = {0};
error = sm_hw_get_if_state( interface->interface_name, &enabled );
if( SM_OKAY != error )
if (SM_SERVICE_DOMAIN_INTERFACE_EVENT_NOT_IN_USE != event)
{
DPRINTFE( "Failed to audit hardware state of interface (%s), "
"error=%s", interface->interface_name,
sm_error_str( error ) );
return( error );
error = sm_hw_get_if_state( interface->interface_name, &enabled );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to audit hardware state of interface (%s), "
"error=%s", interface->interface_name,
sm_error_str( error ) );
return( error );
}
}
switch( event )

23
service-mgmt/sm/src/sm_service_domain_member_table.c
View File

@ -18,7 +18,6 @@
#include "sm_db_service_domain_members.h"
static SmListT* _service_domain_members = NULL;
static SmDbHandleT* _sm_db_handle = NULL;
// ****************************************************************************
// Service Domain Member Table - Read
@ -262,14 +261,6 @@ SmErrorT sm_service_domain_member_table_initialize( void )
_service_domain_members = NULL;
error = sm_db_connect( SM_DATABASE_NAME, &_sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to connect to database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
return( error );
}
error = sm_service_domain_member_table_load();
if( SM_OKAY != error )
{
@ -287,22 +278,8 @@ SmErrorT sm_service_domain_member_table_initialize( void )
// ======================================
SmErrorT sm_service_domain_member_table_finalize( void )
{
SmErrorT error;
SM_LIST_CLEANUP_ALL( _service_domain_members );
if( NULL != _sm_db_handle )
{
error = sm_db_disconnect( _sm_db_handle );
if( SM_OKAY != error )
{
DPRINTFE( "Failed to disconnect from database (%s), error=%s.",
SM_DATABASE_NAME, sm_error_str( error ) );
}
_sm_db_handle = NULL;
}
return( SM_OKAY );
}
// ****************************************************************************

2
service-mgmt/sm/src/sm_service_domain_scheduler.c
View File

@ -1930,7 +1930,7 @@ SmErrorT sm_service_domain_scheduler_initialize( SmDbHandleT* sm_db_handle )
return( error );
}
error = sm_service_domain_weight_initialize( _sm_db_handle );
error = sm_service_domain_weight_initialize();
if( SM_OKAY != error )
{
DPRINTFE( "Failed to intialize service domain weighting, "

8
service-mgmt/sm/src/sm_service_domain_weight.c
View File

@ -16,8 +16,6 @@
#include "sm_db_nodes.h"
#include "sm_service_domain_assignment_table.h"
static SmDbHandleT* _sm_db_handle = NULL;
// ****************************************************************************
// Service Domain Weight - Cleanup
// ===============================
@ -205,10 +203,8 @@ SmErrorT sm_service_domain_weight_apply( char service_domain_name[] )
// ****************************************************************************
// Service Domain Weight - Initialize
// ==================================
SmErrorT sm_service_domain_weight_initialize( SmDbHandleT* sm_db_handle )
SmErrorT sm_service_domain_weight_initialize()
{
_sm_db_handle = sm_db_handle;
return( SM_OKAY );
}
// ****************************************************************************
@ -218,8 +214,6 @@ SmErrorT sm_service_domain_weight_initialize( SmDbHandleT* sm_db_handle )
// ================================
SmErrorT sm_service_domain_weight_finalize( void )
{
_sm_db_handle = NULL;
return( SM_OKAY );
}
// ****************************************************************************

2
service-mgmt/sm/src/sm_service_domain_weight.h
View File

@ -22,7 +22,7 @@ extern SmErrorT sm_service_domain_weight_apply( char service_domain_name[] );
// ****************************************************************************
// Service Domain Weight - Initialize
// ==================================
extern SmErrorT sm_service_domain_weight_initialize( SmDbHandleT* sm_db_handle );
extern SmErrorT sm_service_domain_weight_initialize();
// ****************************************************************************
// ****************************************************************************

2
service-mgmt/sm/src/sm_service_table.c
View File

@ -40,7 +40,6 @@ bool sm_service_clear_failure_state(SmServiceT* service)
bool prev_failure_condition;
prev_failure_condition =
service->recover ||
service->fail_count > 0 ||
service->action_fail_count > 0 ||
service->transition_fail_count > 0 ||
service->status == SM_SERVICE_STATUS_FAILED ||
@ -50,7 +49,6 @@ bool sm_service_clear_failure_state(SmServiceT* service)
if( prev_failure_condition )
{
service->recover = false;
service->fail_count = 0;
service->action_fail_count = 0;
service->transition_fail_count = 0;
service->status = SM_SERVICE_STATUS_NONE;

2
stx-ocf-scripts/centos/build_srpm.data
View File

@ -1,2 +1,2 @@
SRC_DIR="$PKG_BASE/src"
TIS_PATCH_VER=3
TIS_PATCH_VER=PKG_GITREVCOUNT

2
test-requirements.txt
View File

@ -1,3 +1,5 @@
hacking
bashate >= 0.2
PyYAML >= 3.1.0
yamllint >= 0.5.2
bandit;python_version>="3.0"

69
tox.ini
View File

@ -15,6 +15,27 @@ deps = -r{toxinidir}/test-requirements.txt
whitelist_externals = reno
[testenv:linters]
basepython = python3
whitelist_externals = bash
commands =
bash -c "find {toxinidir} \
-not \( -type d -name .?\* -prune \) \
-type f \
-not -name \*~ \
-not -name \*.md \
\( \
-name \*.sh \
-or -not -wholename \*/devstack/files/\* \
-wholename \*/devstack/\* \
\) \
-print0 | xargs -r -n 1 -0 bashate -v -e E* -i E006"
bash -c "find {toxinidir} \
\( -name middleware/io-monitor/recipes-common/io-monitor/io-monitor/io_monitor/test-tools/yaml/* -prune \) \
-o \( -name .tox -prune \) \
-o -type f -name '*.yaml' \
-print0 | xargs -0 yamllint"
[testenv:pylint]
basepython = python2.7
deps = {[testenv]deps}
eventlet
@ -32,24 +53,7 @@ deps = {[testenv]deps}
sqlalchemy-migrate
wsme
pylint
whitelist_externals = bash
commands =
bash -c "find {toxinidir} \
-not \( -type d -name .?\* -prune \) \
-type f \
-not -name \*~ \
-not -name \*.md \
\( \
-name \*.sh \
-or -not -wholename \*/devstack/files/\* \
-wholename \*/devstack/\* \
\) \
-print0 | xargs -0 bashate -v"
bash -c "find {toxinidir} \
\( -name middleware/io-monitor/recipes-common/io-monitor/io-monitor/io_monitor/test-tools/yaml/* -prune \) \
-o \( -name .tox -prune \) \
-o -type f -name '*.yaml' \
-print0 | xargs -0 yamllint"
pylint --rcfile=./pylint.rc --extension-pkg-whitelist=greenlet \
./service-mgmt-api/sm-api/sm_api/ \
./service-mgmt-tools/sm-tools/sm_tools/ \
@ -68,15 +72,24 @@ commands =
# H404: multi line docstring should start without a leading new line
# H405: multi line docstring summary not separated with an empty line
# H501: Do not use locals() for string formatting
# W are warnings
# W504 W504 line break after binary operator
# W605 invalid escape sequence
# - errors -
# E741 ambiguous variable name
# F811 redefinition of unused '<foo>' from line <x>
# F821 undefined name 'e'
# F841 local variable
# F901 'raise NotImplemented' should be 'raise NotImplementedError'
# - bugbear -
# B008 Do not perform calls in argument defaults. The call is performed only once at function definition time.
# B014 Redundant exception types
ignore= E402,
H102,H104,H105,H106,H306,H401,H403,H404,H405,H501,
F811,F821,
B008
W504,W605,
E741,
F811,F821,F841,F901,
B008,B014
# Enable checks which are off by default
# H106 Don’t put vim configuration in source files (off by default). SHOULD BE ENABLED.
# H203 Use assertIs(Not)None to check for None (off by default).
@ -89,9 +102,8 @@ basepython = python3
usedevelop = False
skip_install = True
deps =
hacking
flake8-bugbear<=19.3.0
flake8 < 3.6.0
-r{toxinidir}/test-requirements.txt
flake8-bugbear
commands =
flake8
@ -101,6 +113,7 @@ commands = {posargs}
[testenv:docs]
basepython = python3
install_command = pip install -U {opts} {packages}
deps = -r{toxinidir}/doc/requirements.txt
commands =
rm -rf doc/build
@ -118,15 +131,27 @@ whitelist_externals = rm
[testenv:newnote]
basepython = python3
# Re-use the releasenotes venv
install_command = pip install -U {opts} {packages}
envdir = {toxworkdir}/releasenotes
deps = -r{toxinidir}/doc/requirements.txt
commands = reno new {posargs}
[testenv:api-ref]
basepython = python3
install_command = pip install -U {opts} {packages}
deps =
-r{toxinidir}/doc/requirements.txt
commands =
rm -rf api-ref/build
sphinx-build -W -b html -d api-ref/build/doctrees api-ref/source api-ref/build/html
whitelist_externals = rm
[bandit]
# B411: blacklist Using xmlrpclib
skips = B411
[testenv:bandit]
basepython = python3
description = Bandit code scan for *.py files under config folder
deps = -r{toxinidir}/test-requirements.txt
commands = bandit --ini tox.ini -r {toxinidir}/ -x '**/.tox/**',**/.eggs/** -lll