1fec15cfdc
Modern ARA no longer has ansible in its requirements file so we can remove also passing ANSIBLE_PACKAGE to the installation command. We will already have ansible / ansible-base in the ansible-runtime venv from previous steps in bootstrap-ansible.sh Change-Id: Ic7840d4c64b49c2eccea3dba4798cb5b26b2594c
283 lines
11 KiB
Bash
Executable File
283 lines
11 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
|
|
# Copyright 2016, Rackspace US, Inc.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
# WARNING:
|
|
# This file is use by all OpenStack-Ansible roles for testing purposes.
|
|
# Any changes here will affect all OpenStack-Ansible role repositories
|
|
# with immediate effect.
|
|
|
|
# PURPOSE:
|
|
# This script collects, renames and compresses the logs produced in
|
|
# a role test if the host is in OpenStack-CI.
|
|
|
|
## Vars ----------------------------------------------------------------------
|
|
export WORKING_DIR=${WORKING_DIR:-$(pwd)}
|
|
export RUN_ARA=${RUN_ARA:-false}
|
|
export ARA_REPORT_TYPE=${ARA_REPORT_TYPE:-"database"}
|
|
export TESTING_HOME=${TESTING_HOME:-$HOME}
|
|
export TS=$(date +"%H-%M-%S")
|
|
|
|
export RSYNC_CMD="rsync --archive --copy-links --ignore-errors --quiet --no-perms --no-owner --no-group --whole-file --inplace"
|
|
|
|
# NOTE(cloudnull): This is a very simple list of common directories in /etc we
|
|
# wish to search for when storing gate artifacts. When adding
|
|
# things to this list please alphabetize the entries so it's
|
|
# easy for folks to find and adjust items as needed.
|
|
COMMON_ETC_LOG_NAMES="apt \
|
|
apache2 \
|
|
auditd \
|
|
calico \
|
|
corosync \
|
|
ceph \
|
|
etcd \
|
|
ganesha \
|
|
haproxy \
|
|
httpd \
|
|
memcached \
|
|
mongodb \
|
|
my.cnf \
|
|
mariadb \
|
|
netplan \
|
|
network \
|
|
nginx \
|
|
openstack_deploy \
|
|
pacemaker \
|
|
pip.conf \
|
|
qpid-dispatch \
|
|
rabbitmq \
|
|
repo \
|
|
resolv.conf \
|
|
rsyslog \
|
|
sasl2 \
|
|
sysconfig/network-scripts \
|
|
sysconfig/network \
|
|
systemd \
|
|
uwsgi \
|
|
yum \
|
|
yum.repos.d \
|
|
zypp"
|
|
|
|
COMMON_ETC_LOG_NAMES+=" $(awk -F'os_' '/name.*os_.*/ {print $2}' $(dirname $(readlink -f ${BASH_SOURCE[0]}))/../ansible-role-requirements.yml | tr '\n' ' ')"
|
|
|
|
## Functions -----------------------------------------------------------------
|
|
|
|
function repo_information {
|
|
[[ "${1}" != "host" ]] && lxc_cmd="lxc-attach --name ${1} --" || lxc_cmd=""
|
|
echo "Collecting list of installed packages and enabled repositories for \"${1}\""
|
|
# Redhat package debugging
|
|
if eval sudo ${lxc_cmd} which yum &>/dev/null || eval sudo ${lxc_cmd} which dnf &>/dev/null; then
|
|
# Prefer dnf over yum for CentOS.
|
|
eval sudo ${lxc_cmd} which dnf &>/dev/null && RHT_PKG_MGR='dnf' || RHT_PKG_MGR='yum'
|
|
eval sudo ${lxc_cmd} $RHT_PKG_MGR repolist -v > "${WORKING_DIR}/logs/redhat-rpm-repolist-${1}-${TS}.txt" || true
|
|
eval sudo ${lxc_cmd} $RHT_PKG_MGR list installed > "${WORKING_DIR}/logs/redhat-rpm-list-installed-${1}-${TS}.txt" || true
|
|
|
|
# SUSE package debugging
|
|
elif eval sudo ${lxc_cmd} which zypper &>/dev/null; then
|
|
eval sudo ${lxc_cmd} zypper lr -d > "${WORKING_DIR}/logs/suse-zypper-repolist-${1}-${TS}.txt" || true
|
|
eval sudo ${lxc_cmd} zypper --disable-repositories pa -i > "${WORKING_DIR}/logs/suse-zypper-list-installed-${1}-${TS}.txt" || true
|
|
|
|
# Ubuntu package debugging
|
|
elif eval sudo ${lxc_cmd} which apt-get &> /dev/null; then
|
|
eval sudo ${lxc_cmd} apt-cache policy | grep http | awk '{print $1" "$2" "$3}' | sort -u > "${WORKING_DIR}/logs/ubuntu-apt-repolist-${1}-${TS}.txt" || true
|
|
eval sudo ${lxc_cmd} apt list --installed > "${WORKING_DIR}/logs/ubuntu-apt-list-installed-${1}-${TS}.txt" || true
|
|
|
|
# Gentoo package debugging
|
|
elif eval sudo ${lxc_cmd} which emerge &> /dev/null; then
|
|
# list installed packages
|
|
eval sudo ${lxc_cmd} equery list "*" > "${WORKING_DIR}/logs/gentoo-portage-list-installed-${1}-${TS}.txt" || true
|
|
# list only packages called for install (not dependancies)
|
|
eval sudo ${lxc_cmd} cat /var/lib/portage/world > "${WORKING_DIR}/logs/gentoo-portage-list-manual-installed-${1}-${TS}.txt" || true
|
|
fi
|
|
|
|
}
|
|
|
|
function store_artifacts {
|
|
# Store known artifacts only if they exist. If the target directory does
|
|
# exist, it will be created.
|
|
# USAGE: store_artifacts /src/to/artifacts /path/to/store
|
|
if sudo test -e "${1}"; then
|
|
if [[ ! -d "${2}" ]]; then
|
|
mkdir -vp "${2}"
|
|
fi
|
|
echo "Running artifact sync for \"${1}\" to \"${2}\""
|
|
sudo ${RSYNC_CMD} ${1} ${2} || true
|
|
fi
|
|
}
|
|
|
|
function store_journal_artifacts {
|
|
# Store lines from a known unit's journal as a plain-text log file.
|
|
# USAGE: store_journal_artifacts UNIT_TO_MATCH /path/to/store
|
|
if [ $? == 0 ]; then
|
|
if [[ ! -d "${2}" ]]; then
|
|
mkdir -vp "${2}"
|
|
fi
|
|
if [[ ${3:-false} != false ]]; then
|
|
if [[ -f "${3}/system.journal" ]]; then
|
|
SYSTEMD_UNITS=$(sudo journalctl --file="${3}/system.journal" -F _SYSTEMD_UNIT | grep "${service}")
|
|
for service_unit in $(echo -e "${SYSTEMD_UNITS}"); do
|
|
echo "Pulling journal for ${service_unit}"
|
|
sudo journalctl --file="${3}/system.journal" \
|
|
--unit="${service_unit}" | sudo tee "${2}/${service_unit}.journal-${TS}.log" &>/dev/null
|
|
done
|
|
fi
|
|
else
|
|
SYSTEMD_UNITS=$(sudo journalctl --output=json-pretty -F _SYSTEMD_UNIT | grep "${service}")
|
|
for service_unit in $(echo -e "${SYSTEMD_UNITS}"); do
|
|
echo "Pulling journal for ${service_unit}"
|
|
sudo journalctl --unit="${service_unit}" | sudo tee "${2}/${service_unit}.journal-${TS}.log" &>/dev/null
|
|
done
|
|
fi
|
|
fi
|
|
}
|
|
|
|
function find_files {
|
|
find "${WORKING_DIR}/logs/" -type f \
|
|
! -name "*.gz" \
|
|
! -name '*.html' \
|
|
! -name '*.subunit' \
|
|
! -name "*.journal" \
|
|
! -name 'ansible.sqlite' | egrep -v 'stackviz|ara-report'
|
|
}
|
|
|
|
function rename_files {
|
|
find_files |\
|
|
while read filename; do \
|
|
mv ${filename} ${filename}.txt || echo "WARNING: Could not rename ${filename}"; \
|
|
done
|
|
}
|
|
|
|
## Main ----------------------------------------------------------------------
|
|
|
|
echo "#### BEGIN LOG COLLECTION ###"
|
|
|
|
mkdir -vp "${WORKING_DIR}/logs"
|
|
|
|
# Gather basic logs
|
|
store_artifacts /openstack/log/ansible-logging/ "${WORKING_DIR}/logs/ansible"
|
|
store_artifacts /openstack/log/ "${WORKING_DIR}/logs/openstack"
|
|
store_artifacts /var/log/ "${WORKING_DIR}/logs/host"
|
|
|
|
# Build the ARA static html report if required
|
|
if [[ "$ARA_REPORT_TYPE" == "html" ]]; then
|
|
echo "Generating ARA static html report."
|
|
/opt/ansible-runtime/bin/ara-manage generate "${WORKING_DIR}/logs/ara-report"
|
|
fi
|
|
|
|
# Store the ara sqlite database in the openstack-ci expected path
|
|
store_artifacts "${TESTING_HOME}/.ara/ansible.sqlite" "${WORKING_DIR}/logs/ara-report/"
|
|
|
|
# Store netstat report
|
|
store_artifacts /tmp/listening_port_report.txt "${WORKING_DIR}/logs/host"
|
|
|
|
# Copy the repo os-releases *.txt files
|
|
# container path
|
|
store_artifacts /openstack/*repo*/repo/os-releases/*/*/*.txt "${WORKING_DIR}/repo"
|
|
|
|
# metal path
|
|
store_artifacts /var/www/repo/os-releases/*/*/*.txt "${WORKING_DIR}/repo"
|
|
|
|
# Verify the integrity of the journal files but do not fail if one of them is not usable
|
|
echo "Verifying journal files consistency..."
|
|
find /var/log/journal/ -type f -name "*.journal" -exec bash -c 'sudo journalctl --file={} --verify || true' \;
|
|
|
|
|
|
# Gather host etc artifacts
|
|
PIDS=()
|
|
for service in ${COMMON_ETC_LOG_NAMES}; do
|
|
echo "Running collection for service ${service}"
|
|
store_artifacts "/etc/${service}" "${WORKING_DIR}/logs/etc/host/" &
|
|
pid=$!
|
|
PIDS[${pid}]=${pid}
|
|
store_journal_artifacts "${service}" "${WORKING_DIR}/logs/host" &
|
|
pid=$!
|
|
PIDS[${pid}]=${pid}
|
|
done
|
|
echo "Waiting for host collection jobs to finish"
|
|
for job_pid in ${!PIDS[@]}; do
|
|
wait ${PIDS[$job_pid]} || exit 99
|
|
done
|
|
|
|
|
|
# Gather container etc artifacts
|
|
if which lxc-ls &> /dev/null; then
|
|
for CONTAINER_NAME in $(sudo lxc-ls -1); do
|
|
CONTAINER_PID=$(sudo lxc-info -p -n ${CONTAINER_NAME} | awk '{print $2}')
|
|
ETC_DIR="/proc/${CONTAINER_PID}/root/etc"
|
|
MACHINE_ID="$(sudo cat ${ETC_DIR}/machine-id)"
|
|
LOG_DIR="/proc/${CONTAINER_PID}/root/var/log"
|
|
JOURNAL_DIR="/proc/${CONTAINER_PID}/root/run/log/journal/${MACHINE_ID}"
|
|
repo_information ${CONTAINER_NAME}
|
|
PIDS=()
|
|
for service in ${COMMON_ETC_LOG_NAMES}; do
|
|
echo "Running in container collection for service ${service}"
|
|
store_artifacts ${ETC_DIR}/${service} "${WORKING_DIR}/logs/etc/openstack/${CONTAINER_NAME}/" &
|
|
pid=$!
|
|
PIDS[${pid}]=${pid}
|
|
store_artifacts ${LOG_DIR}/${service} "${WORKING_DIR}/logs/openstack/${CONTAINER_NAME}/" &
|
|
pid=$!
|
|
PIDS[${pid}]=${pid}
|
|
store_journal_artifacts ${service} "${WORKING_DIR}/logs/openstack/${CONTAINER_NAME}" "${JOURNAL_DIR}" &
|
|
pid=$!
|
|
PIDS[${pid}]=${pid}
|
|
done
|
|
echo "Waiting for container collection jobs for ${CONTAINER_NAME} to finish"
|
|
for job_pid in ${!PIDS[@]}; do
|
|
wait ${PIDS[$job_pid]} || exit 99
|
|
done
|
|
done
|
|
fi
|
|
|
|
|
|
# Rename all files gathered to have a .txt suffix so that the compressed
|
|
# files are viewable via a web browser in OpenStack-CI.
|
|
rename_files
|
|
|
|
# If we could not find ARA, assume it was not installed
|
|
# and skip all the related activities.
|
|
if [ "${RUN_ARA}" = true ]; then
|
|
# Generate the ARA subunit report so that the
|
|
# results reflect in OpenStack-Health
|
|
mkdir -vp "${WORKING_DIR}/logs/ara-data"
|
|
echo "Generating ARA report subunit report."
|
|
/opt/ansible-runtime/bin/ara generate subunit "${WORKING_DIR}/logs/ara-data/testrepository.subunit" || true
|
|
fi
|
|
|
|
# Get a dmesg output so we can look for kernel failures
|
|
dmesg > "${WORKING_DIR}/logs/dmesg-${TS}.txt" || true
|
|
|
|
# Collect job environment
|
|
env > "${WORKING_DIR}/logs/environment-${TS}.txt" || true
|
|
|
|
repo_information host
|
|
|
|
# Record the active interface configs
|
|
if which ethtool &> /dev/null; then
|
|
for interface in $(ip -o link | awk -F':' '{print $2}' | sed 's/@.*//g'); do
|
|
echo "ethtool -k ${interface}"
|
|
ethtool -k ${interface} > "${WORKING_DIR}/logs/ethtool-${interface}-${TS}-cfg.txt" || true
|
|
done
|
|
else
|
|
echo "No ethtool available" | tee -a "${WORKING_DIR}/logs/ethtool-${TS}-${interface}-cfg.txt"
|
|
fi
|
|
|
|
# Ensure that the files are readable by all users, including the non-root
|
|
# OpenStack-CI jenkins user.
|
|
sudo chmod -R ugo+rX "${WORKING_DIR}/logs"
|
|
sudo chown -R $(whoami) "${WORKING_DIR}/logs"
|
|
|
|
echo "#### END LOG COLLECTION ###"
|
|
|