Let admins add a region without melting their cluster
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
openstack-manuals |
Invalid
|
Medium
|
Unassigned |
Bug Description
https:/
commit 6d77c379bd8f51e
Author: Samuel Merritt <email address hidden>
Date: Tue Aug 19 16:44:56 2014 -0700
Let admins add a region without melting their cluster
Prior to this commit, swift-ring-builder would place partitions on
devices by first going for maximal dispersion and breaking ties with
device weight. This commit flips the order so that device weight
trumps dispersion.
Note: if your ring can be balanced, you won't see a behavior
change. It's only when device weights and maximal-dispersion come into
conflict that this commit changes anything.
Example: a cluster with two regions. Region 1 has a combined weight of
1000, while region 2 has a combined weight of only 400. The ring has 3
replicas and 2^16 partitions.
Prior to this commit, the balance would look like so:
Region 1: 2 * 2^16 partitions
Region 2: 2^16 partitions
After this commit, the balance will be:
Region 1: 10/14 * 2^16 partitions (more than before)
Region 2: 4/14 * 2^16 partitions (fewer than before)
One consequence of this is that some partitions will not have a
replica in region 2, since it's not big enough to hold all of them.
This way, a cluster operator can add a new region to a single-region
cluster in a gradual fashion so as not to destroy their WAN link with
replication traffic. As device weights are increased in the second
region, more replicas will shift over to it. Once its weight is half
that of the first region's, every partition will have a replica there.
DocImpact
Change-Id: I945abcc4a2917b
commit 8fecf490fe3e1be
Author: Samuel Merritt <email address hidden>
Date: Tue Aug 19 14:50:44 2014 -0700
Respect device weights when adding replicas
Previously, any new (partition, replica) pairs created by adding
replicas were spread evenly across the whole ring. Now it respects
device weights, so more (partition, replica) pairs are placed on
higher-weight devices.
Note that this only affects partition assignment *when the replica
count is increased*. Normal adding of disks without changing the
replica count is, and was, just fine.
The cause was that the devices' parts_wanted values weren't being
updated when the replica count went up. Since adding replicas makes
more things to assign, each device's desired share should have gone
up, but it didn't. This appears to be a simple oversight on the part
of the original author (me).
Change-Id: Idab14be90fab24
commit 8526a071909f328
Author: paul luse <email address hidden>
Date: Tue Aug 19 07:10:06 2014 -0700
Fix sporadic false failure in xprofile unit test code (master)
Same fix as is going through on feature/ec, fixed it there first
as it was happening often and no point in waiting until EC is
complete before getting it over to master...
Appears that what's been happening on feature/ec lately
with a middleware failure has to do with a hardcoded PID in
the test code itself causing a profile file to exist when
its not expected to by the test.
Test code used a PID of 135 and based on how get_logfiles()
is written, any real PID that starts with 135 will cause a
false failure in test_call(). This can be seen via inspection
and confirmed in logfiles where all captured assertions show
a profile filename beginning with 135. Tried getting smarter
about choosing a fake PID (int) but then decided it was 100%
safe to use 'ABC' for this test since that'll never show up!
Change-Id: I958f1525c2727b
commit 75a329c7f54fc3b
Author: John Dickinson <email address hidden>
Date: Sun Aug 17 21:08:57 2014 -0700
add the account management config values to swift info
added account_autocreate and allow_account_
the /info endpoint
Change-Id: I4b239c9cefb728
commit 0abd2cba035f9f9
Author: David Goetz <email address hidden>
Date: Mon Aug 11 15:08:18 2014 -0700
Shard expiring object container
All the expiring objects for a given X-Delete-At are funnelled into the
same expiring object container- this can act as a bottleneck.
Change-Id: I288a177a7ae3e2
commit a0e0014159c0e12
Author: David Goetz <email address hidden>
Date: Mon Aug 11 09:43:13 2014 -0700
Sleep for longer at a time in lock_path.
When lock_path is called and the lock goes for the whole 10 seconds,
the flock is called 1000 times. With this patch, the short 0.01 sleep
is used for the first 1% of the total lock time and then 1% of the
total lock time is used.
Change-Id: Ibed6bdb49bddcd
commit d8b3e16c038eac0
Author: Alistair Coles <email address hidden>
Date: Mon Aug 4 17:22:08 2014 +0100
Update tempurl docstring with methods config option
Adds a description of the methods config option to the
docstring.
Also fixes description of swift.source value and a
couple of other typos.
Change-Id: If3489087df0205
commit e114a628dcaf59e
Author: Clay Gerrard <email address hidden>
Date: Tue Jul 1 11:23:48 2014 -0700
Remove fake _get_part and use the real thing
The get_part method is fast and stable given a consistent hash path
suffix/prefix, so there's no absolute requirement for the fake
implementation other than convenience. OTOH, removing the fake
implementation and fixing the tests that were relying on it should make
it easier to write better tests going forward and harder to hide bugs
that don't show up when using the fakes.
There may be some overhead when writing new tests that use the ring if
you're making assertions on partitions or paths, but with a part power
of zero it's normally trivially obvious when a 1 needs to be a 0 or vice
versa. Or you can just drop the assertions around the parts you were
faking anyway.
Change-Id: I8bfc388a04eff6
commit ba38ba5df54b8eb
Author: Christian Schwede <email address hidden>
Date: Fri Jun 13 10:33:03 2014 +0000
Fix object auditor recon and logging
1. Nothing is logged until at least one audit needs more than log_time.
If the audit runtime never exceeds this value (which is 3600 seconds
by default) nothing is logged and the recon entry is never updated.
It happens especially on very fast disks with a low usage and/or if
only a few disks are audited (for example, using the --devices
parameter on the command line).
This patch changes this to log and update the recon cache entry
at least one time after the first device audit.
2. If device_dirs is set the recon entry will be deleted after all
devices are audited.
Change-Id: Ifa504d21389b3a
Changed in openstack-manuals: | |
milestone: | none → juno |
status: | New → Triaged |
importance: | Undecided → Medium |
Changed in openstack-manuals: | |
status: | Triaged → Invalid |
milestone: | juno → none |