Greg Farnum [Thu, 29 Aug 2013 22:26:08 +0000 (15:26 -0700)]
workunits: add a test for caching redirects
This may need to change since it exploits some of the loose
consistency we currently have with caching pools, but for now
it checks that the Objecter does what we want.
Greg Farnum [Thu, 29 Aug 2013 20:52:35 +0000 (13:52 -0700)]
Objecter: be careful about precalculated pgids
The only current user of the precalc_pgid field is list_objects. That's
fine, but we don't want new users to inadvertently appear and somehow
break the caching/tiering stuff by forcing us to go to the base pool
when we should be talking to somebody else. Add an assert to catch
these cases.
Greg Farnum [Thu, 29 Aug 2013 20:08:03 +0000 (13:08 -0700)]
Objecter: rename Op::oloc -> Op::base_oloc
We want to be able to target other pools for caching and tiering, so
we need to take an oloc from the client and translate it into an
actual target. Rename oloc to base_oloc to make clear which one it is.
Sage Weil [Mon, 26 Aug 2013 22:59:54 +0000 (15:59 -0700)]
osd_types: add pg_pool_t cache-related fields
We add fields sufficient to specify
* many pools have a tiering relationship with pool foo
* pool foo is a tier pool for pool bar
* the tiering relationship between foo and bar is specified
by cache_mode
* client reads and writes for pool foo should be directed to
pools bar and baz, respectively (where probably, but not
necessarily, baz == bar or baz == foo).
This lets us specify very sophisticated caching policies on
the server side that all clients going forward can handle
simply by directing the messages as the read_tier and write_tier
flags, and the (not-yet-implemented) redirect replies
from OSDs, specify.
Signed-off-by: Sage Weil <sage@inktank.com> Signed-off-by: Greg Farnum <greg@inktank.com>
Greg Farnum [Wed, 28 Aug 2013 00:24:24 +0000 (17:24 -0700)]
ReplicatedPG: add OpContext::user_at_version
Set this up with the existing at_version member, but only increase
it for user_modify ops. Use this when logging the PG's user_version. In
order to maintain compatibility with old clients on classic pools, we
force user_version to follow at_version whenever it's updated.
Now that we have and are maintaining this PG user version, use it
for the user version on ops that get ENOENT back, when short-circuiting
replies as part of reply_op_error()[1], or when replying to repops
in eval_repop; further use it for the cls_current_version() function. This
is a small semantic change for that function, as previously it would
generally return the same value as the user would get sent back via
MOSDOpReply -- but I don't think it was something you could count on.
We now define it as being the user version of the PG at the start of the
op, and as a bonus it is defined even for read ops (the at_version is
only filled in on write operations).
[1]: We tweak PGLog to make it easier to retrieve both user and PG versions.
Greg Farnum [Tue, 27 Aug 2013 18:02:44 +0000 (11:02 -0700)]
MOSDOpReply: add enough fields to be backwards compatible.
The system we've been building up works out very nicely for new clients,
but they could not have interoperated with old clients that were only
referring to our replay_version. In order to deal with this, we add
a bad_replay_version to MOSDOpReply which is encoded where we used
to encode replay_version. bad_replay_version will follow the same semantics
as reassert_version used to (except that it is filled in on reads), but
is not accessible to new clients, who can see only our properly-controlled
replay_version and user_version. This will let old and new clients
interoperate correctly when communicating about watches, etc.
Greg Farnum [Wed, 28 Aug 2013 00:14:56 +0000 (17:14 -0700)]
osd: actually fill in user_version in pg_log_entry_t
We now require it when creating a pg_log_entry_t. The user_version
is the version which info.last_user_version should be set to
after the transaction is applied, which for everything except for
a user-modify op is going to be the version it was already at.
For now we are filling in the user-modify op's changing user_version
to be ctx->at_version.version
Greg Farnum [Wed, 21 Aug 2013 18:26:28 +0000 (11:26 -0700)]
osd: add last_user_version to pg_info_t
We add a corresponding user_version to pg_log_entry_t, and the logic
to assign from one to the other and to recover last_user_version from
a master's log. We aren't yet setting it to anything, though.
ctx->new_obs.oi.user_version is initialized to ctx->obs.oi.user_version,
and for read ops it won't be changed. That means
reply_user_version == ctx->new_obs.oi.user_version in all cases, which
means we don't want it.
Greg Farnum [Wed, 21 Aug 2013 00:13:53 +0000 (17:13 -0700)]
osd: switch object_info_t::user_version to be a version_t
We never expose the full eversion_t data to users, and do not want to.
However, we pull some tricks in the encode/decode functions to avoid
having to change the object_info_t disk format for this change.
When we can break compatibility, we should simplify this.
Greg Farnum [Tue, 20 Aug 2013 23:22:27 +0000 (16:22 -0700)]
ReplicatedPG: Fill in the MOSDOpReply's user_version
As part of this, rename OpContext::reply_version->reply_user_version.
The semantics that necessitate the reply_version are only for user versions,
so rename it for clarity. Then use the reply_user_version in
set_user_version() (if the op succeeded).
For now we use the PG version for ENOENT (preserving the previous
semantics), but that will get changed to the pg's user_version soon
as well.
Greg Farnum [Tue, 20 Aug 2013 21:21:04 +0000 (14:21 -0700)]
Objecter: librados: mass switch from eversion_t to version_t
There are a lot of pointers throughout our request infrastructure used solely
for exporting the version to users. The interfaces we actually expose only
provide a uint64_t (leaving off eversion_t's epoch), and that's all we're
going to maintain in our new user_version scheme, so don't pretend we'll
have more in our internal interfaces.
I audited this pretty carefully; in particular:
Op::objver is only used for passing data back to users via the calling
functions IoCtxImpl::last_objver, etc
IoCtxImpl::last_objver is used only for the set_sync_op_version() call, which
provides data only for the uint64_t get_last_version() and
rados_get_last_version() calls.
AioCompletionImpl::objver is used only for the uint64_t get_version() call.
LingerOp::pobjver is used only for referencing things that are now version_t.
Greg Farnum [Thu, 22 Aug 2013 21:54:19 +0000 (14:54 -0700)]
ReplicatedPG: do not do a redundant set of ctx->new_obs.oi.version
We set this in the if below for writes, and for reads it doesn't need to
be updated (and isn't). Remove the confusing double-set so future code
inspectors don't get concerned there's a bug like I did.
Greg Farnum [Mon, 26 Aug 2013 21:38:30 +0000 (14:38 -0700)]
ReplicatedPG: remove long-dead branch
This was confusing the heck out of me when trying to figure out
why I was hitting an assert. So replace the if-else block with
a more appropriate assert and don't include any misleading calls
to prepare_transaction() from sub_op_modify().
We have been returning the object's "user version" and using that
for replay, but that is in fact incorrect. In preparation for fixing
up the user version semantics, rename get_version to get_replay_version
and set_version to set_replay_version.
Loic Dachary [Tue, 27 Aug 2013 11:58:33 +0000 (13:58 +0200)]
common: move SharedPtrRegistry test after t.join
The thread created to test SharedPtrRegistry race conditions updates a
value ( ptr ) that is tested by the main gtest thread but is not
protected by a lock. Instead of adding a lock, the main thread tests
the value after pthread_join() on the child thread.
mon: DataHealthService: monitor backing store's size and report it
If the store's size grows beyond what we believe to be reasonable, we must
let the user know that something fishy may be going on. This intends to
act as an early warning system for monitors suffering from leveldb
compaction issues. However, if the monitor's store is just growing a lot
due to normal cluster behaviour, we made sure that the warning threshold
is adjustable by tuning 'mon_leveldb_size_warn' (defaulting to 40GB).
Fixes: #5909 Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
Dan Mick [Fri, 23 Aug 2013 00:30:24 +0000 (17:30 -0700)]
ceph_rest_api.py: create own default for log_file
common/config thinks the default log_file for non-daemons should be "".
Override that so that the default is
/var/log/ceph/{cluster}-{name}.{pid}.log
since ceph-rest-api is more of a daemon than a client.
Fixes: #6099
Backport: dumpling Signed-off-by: Dan Mick <dan.mick@inktank.com>
David Disseldorp [Mon, 29 Jul 2013 15:05:44 +0000 (17:05 +0200)]
mds: remove waiting lock before merging with neighbours
CephFS currently deadlocks under CTDB's ping_pong POSIX locking test
when run concurrently on multiple nodes.
The deadlock is caused by failed removal of a waiting_locks entry when
the waiting lock is merged with an existing lock, e.g:
Note that the waiting 4116@1:1 lock entry is merged with the existing
4116@0:1 held lock to become a 4116@0:2 held lock. However, the now
handled 4116@1:1 waiting_locks entry remains.
When handling a lock request, the MDS calls adjust_locks() to merge
the new lock with available neighbours. If the new lock is merged,
then the waiting_locks entry is not located in the subsequent
remove_waiting() call because adjust_locks changed the new lock to
include the old locks.
This fix ensures that the waiting_locks entry is removed prior to
modification during merge.
Signed-off-by: David Disseldorp <ddiss@suse.de> Reviewed-by: Greg Farnum <greg@inktank.com>
os: KeyValueDB: expose interface to obtain estimated store size
On LevelDBStore, instead of using leveldb's GetApproximateSizes() function,
we will instead assess what's the store's raw size from the contents of
the store dir (this means .sst's, .log's, etc). The reason behind this
approach is that GetApproximateSizes() would expect us to provide a range
of keys for which to obtain an approximate size; on the other hand, what we
really want is to obtain the size of the store -- not the size of the
data (besides, with the compaction issues we've been seeing, we wonder
how reliable such approximation would be).
Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
Sage Weil [Thu, 22 Aug 2013 22:54:48 +0000 (15:54 -0700)]
mon/Paxos: fix another uncommitted value corner case
It is possible that we begin the paxos recovery with an uncommitted
value for, say, commit 100. During last/collect we discover 100 has been
committed already. But also, another node provides an uncommitted value
for 101 with the same pn. Currently, we refuse to learn it, because the
pn is not strictly > than our current uncommitted pn... even though it is
the next last_committed+1 value that we need.
There are two possible fixes here:
- make this a >= as we can accept newer values from the same pn.
- discard our uncommitted value metadata when we commit the value.
Let's do both!
Fixes: #6090 Signed-off-by: Sage Weil <sage@inktank.com>
Yehuda Sadeh [Mon, 19 Aug 2013 23:56:27 +0000 (16:56 -0700)]
rgw: bucket meta remove don't overwrite entry point first
Fixes: #6056
When removing a bucket metadata entry we first unlink the bucket
and then we remove the bucket entrypoint object. Originally
when unlinking the bucket we first overwrote the bucket entrypoint
entry marking it as 'unlinked'. However, this is not really needed
as we're just about to remove it. The original version triggered
a bug, as we needed to propagate the new header version first (which
we didn't do, so the subsequent bucket removal failed).
Sandon Van Ness [Fri, 23 Aug 2013 02:44:40 +0000 (19:44 -0700)]
QA: Compile fsstress if missing on machine.
Some distro's have a lack of ltp-kernel packages and all we need is
fstress. This just modified the shell script to download/compile
fstress from source and copy it to the right location if it doesn't
currently exist where it is expected. It is a very small/quick
compile and currently only SLES and debian do not have it already.
Reviewed-by: Sage Weil <sage@inktank.com> Signed-off-by: Sandon Van Ness <sandon@inktank.com>
Sandon Van Ness [Fri, 23 Aug 2013 02:44:40 +0000 (19:44 -0700)]
QA: Compile fsstress if missing on machine.
Some distro's have a lack of ltp-kernel packages and all we need is
fstress. This just modified the shell script to download/compile
fstress from source and copy it to the right location if it doesn't
currently exist where it is expected. It is a very small/quick
compile and currently only SLES and debian do not have it already.
Reviewed-by: Sage Weil <sage@inktank.com> Signed-off-by: Sandon Van Ness <sandon@inktank.com>
Gary Lowell [Thu, 22 Aug 2013 18:07:16 +0000 (11:07 -0700)]
ceph.spec.in: Don't invoke debug_package macro on centos.
If the redhat-rpm-config package is installed, the debuginfo rpms will
be built by default. The build will fail when the package installed
and the specfile also invokes the macro.
Signed-off-by: Gary Lowell <gary.lowell@inktank.com>
projects / ceph.git / log