The log_statsd_host value can now be an IPv6 address or a hostname
which only resolves to an IPv6 address. In both cases, the new
behavior is to use an AF_INET6 socket on which .sendto() is called
with the originally-configured hostname (or IP). This means the
Swift process is not caching a DNS resolution for the lifetime of
the process (a good thing).
If a hostname resolves to both an IPv6 or IPv4 address, an AF_INET
socket is used (i.e. only the IPv4 address will receive the UDP
packet).
The old behavior is preserved: any invalid IP address literals and
failures in DNS resolution or actual StatsD packet sending do not
halt the process or bubble up; they are caught, logged, and
otherwise ignored.
Change-Id: Ibddddcf140e2e69b08edf3feed3e9a5fa17307cf
liberasurecode_rs_vand is build-in liberasurecode, so you don't need
another depedency libjerasure2.
liberasurecode_rs_vand is supported by pyeclib from 1.0.8
version, so bumping version up.
Closes-Bug: #1534325
Change-Id: If2d96875694df8fd48c5278395859aaa165cb566
If not explicitly configured the versioned_writes middleware
should be auto-inserted in the pipeline after slo and dlo, which
is where the versioned_writes filter section's comments say it
should be in proxy-server.conf-sample. At the moment it can end up
being placed ahead of slo and dlo if they have been explicitly
configured, which results in the linked bug manifesting.
Closes-Bug: #1537042
Change-Id: I6ac95a331f4ef0d4887311940acc6f8bc00fb4eb
I've found that given a sufficiently bad replica2part2dev table we can
accidently not entirely fix palcement when more than two replicas of a
part are assigned to the duplicate devices.
It shows up most on > 3 replica rings when you have two *different*
devices both holding two replicas. But you can see it on a three
replica ring when all three replicas are assigned to the same device.
Change-Id: Ieb213c1a259815a2ed657291242919cda568c7b5
Running functional tests in the in-process mode uses
the default value for proxy-server object_post_as_copy,
which is True. This patch adds support for an environment
variable to change this value to False so that the fast-post
mode is tested when running in-process functional tests.
The patch also adds a new tox environment func-in-process-fast-post
which forces in-process functional testing with
object_post_as_copy=False.
The motivation for this change, apart from enabling configurable
local testing, is to put support in place for an upstream CI job
that will actually functionally test the fast-post mode, which is
otherwise only covered by unit tests.
There are currently two gate jobs that run the functional tests:
- gate-swift-dsvm-functional runs the tests *twice* against a
devstack swift service, once using tempauth and once using
keystoneauth. The devstack swift service uses the default
object_post_as_copy=True.
- gate-swift-tox-func runs the func tests in in-process mode
which also uses tempauth and object_post_as_copy=True. This
duplicates one of the config scenarios above.
With this change either the gate-swift-tox-func job or a new job
could run the functional tests using object_post_as_copy=False.
Change-Id: Ia37f6df1dc38e44ef7404fbf0a52f6fc22fae0c2
Including a * in the param name causes a sphinx warning,
messes up the html formatting, and strictly the * isn't part of
the param name.
Change-Id: I6be6a7aa2a371eaae9f8e501435838c1e15f4a1c
In the section of General Service Tuning, there is a sentence
"Our Storage servers all run together on the same servers".
The word "servers" is used twice with different meanings, which
would be a little confusing. This patch makes it more understandable.
Change-Id: I6d806456aeb7d0ecf297c04dacbd271b94a0425c
Closes-Bug: #1526697
Make the result of Timestamp(x) != Timestamp(x) be False.
In python 2.7 this requires the __ne__ method to be defined [1].
"The truth of x==y does not imply that x!=y is false." The
functools.total_ordering decorator does not autocreate a __ne__
method.
In python 3 the __ne__ method is not required [2]. "By default,
__ne__() delegates to __eq__() and inverts the result".
This patch puts back the __ne__ method removed in [3]. Whilst no tests
fail on master with python2.7, they do on this patch [4] and it seems
dangerous to have this absurd behaviour lurking.
[1] https://docs.python.org/2/reference/datamodel.html#object.__ne__
[2] https://docs.python.org/3.4/reference/datamodel.html#object.__ne__
[3] Change-Id: Id26777ac2c780316ff10ef7d954c48cc1fd480b5
[4] Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
Change-Id: I01fbfa310df3c74390f8e8c2e9ffff81bbf05e47
These errors are producing lintian warnings, so fixing them
helps having less errors when checking for Debian packages.
Change-Id: Iff99a8d5f2276515f42d758d110a43cae757db28
This patch provides unit tests to check responses for various situation
in three/four replicas env.
Change-Id: I1af48728b28da9780a2a78f3110e1c7487047ff8
If "Permission Denied" has happen in NamedTemporaryFile function in
dump_recon_cache method, swift will log a message of reference to a variable
without assignment and not log a message of "Permission Denied".
This patch fixes the handling and add an unit test.
Co-Authored-By: Kota Tsuyuzaki <tsuyuzaki.kota@lab.ntt.co.jp>
Change-Id: Iafdd94905e9e9c81f5966a923324b50c18fcf592
Based on experience using handoffs_first and feedback from other
operators it has become clear that handoffs_first is only used during
periods of problematic cluster behavior (e.g. full disks) when
replication attempts are failing to quickly drain off the partitions
from the nodes which they have been rebalanced from.
In order to focus on the most important work (getting handoff partitions
off the node) handoffs_first mode will abort the current replication
sweep before attempting any primary suffix syncing if any of the handoff
partitions were not removed for any reason - and start over with
replication of handoffs jobs as the highest priority.
Note that handoffs_first being enabled will emit a warning on start up,
even if no handoff jobs fail, because of the negative impact it can have
during normal operations by dog piling on a node that was temporarily
unavailable.
Change-Id: Ia324728d42c606e2f9e7d29b4ab5fcbff6e47aea
It feels silly, and we don't do it for any of the other headers in
headers_to_container_info.
While we're at it, clean up a stray '
Change-Id: I0745038cc3832a77d064e515c37cacbdcb97c4d9
Related-Change: Iea3d06de80210e9e504e296d4572583d7ffabeac
