Abstract out how writeback is done with a WritebackHandler object.
For RBD caching, this will be done by librados, but the Client uses
the Objecter directly.
This also requires different locks, since librbd does not have access
to the lock the underlying Objecter uses. Thus, both lock and the
writeback handler are parameters of the ObjectCacher constructor.
Sage Weil [Thu, 5 Apr 2012 22:55:50 +0000 (15:55 -0700)]
librados: do aio callbacks in async thread
Call user completions in an async thread. This allows callers to call back
into librados from the callback, and allows them to take locks in their
callbacks that they hold when queuing requests (making their life much
easier).
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
DeterministicOpSequence: writing to object being cloned in the same tx.
We write to the object being cloned prior to the clone to ensure we are
cloning a valid range of bytes.
The write and the clone were being done in two distinct transactions,
which would trigger a diff mismatch if a failure happened to occur within
the write tx. By not differentiating the transactions when building a
pristine copy, we were executing one more transaction (the clone_range one)
than the transactions that were executed in the failed filestore, thus
triggering the mismatch (one more object in the pristine filestore than on
the failed filestore).
Now we issue a single transaction, containing both the clone_range()
preceded by a write().
Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
Sage Weil [Thu, 12 Apr 2012 17:40:08 +0000 (10:40 -0700)]
filestore: name internally
We need to allow the perfcounter name to be controlled so that we can have
two instances of FileStore in the same process that don't step on each
other. Default to 'filestore'.
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
test_idempotent_sequence: Generate a reproducible sequence of txs.
With this test we aim at reproducing the same sequence of transactions
as long as we are provided with the same seed between runs.
We also allow failures to be injected onto the filestore if the
--filestore-kill-at <VAL> argument is passed, and we provide verification
when --test-verify-at <VAL> is provided.
Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
VerifyFileStore: Check if two FileStore's match after applying a set of operations.
With DeterministicOpSequence we are able to reproduce exactly the same
sequence of operations, over and over. However, if the filestore fails
(e.g., because we injected a failure), we want to check if it is kept
consistent after replaying its journal.
With VerifyFileStore, which extends DeterministicOpSequence, we are able
to bring a brand new filestore to the state the failed filestore would
reach had it not failed. We can then compare to check if the failure
introduced inconsistencies after replaying the journal.
(This is still work in progress and not fully functional)
Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
TestFileStoreState: Represent a FileStore's state to be used by tests.
Instead of having each test creating the same representation of a
FileStore's state, with a map/set of collections and objects, as well as
multiple init() functions for each test that are in all similar in
nature, provide this in a single class that can be inheritted by test
classes.
Sage Weil [Sat, 14 Apr 2012 00:11:54 +0000 (17:11 -0700)]
filestore: two-phase guard
For certain operations (collection_add) we need a two-phase guard, and an
"in-progress" state.
- before exposing an object in a new location, we need to mark it so that
old operations affecting the target name don't touch the new object.
- can't just set the guard before starting or else we can't distinguish
between a collection_add that was in-progress and one that happend a
long time ago.
We may need the same for collection_rename().
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
Sage Weil [Sat, 14 Apr 2012 00:17:36 +0000 (17:17 -0700)]
filestore: simple failure injections via --filestore-kill-at <n>
This will make filestore suicide (_exit(1)) on the n'th potential failure
call site. We can potentially fail:
- before a transaction
- between each op
- at the end
Additionally, we instrument the guards:
- before/after/inside _set_replay_guard
- between significant steps of callers of _set_replay_guard
All instrumentation points are inside _do_transactions(), so if everything
is done in a single sequencer (or from a single thread) the failure
point is deterministic.
That said, use an atomic so we will still reliably fail (at some point)
when there are multiple filestore threads in action.
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
Sage Weil [Fri, 13 Apr 2012 18:30:49 +0000 (11:30 -0700)]
filestore: replay collection_move using add+remove
This approximates the buggy collection_move. It is still buggy. It is
only there to replay old journals.
Rip out buggy (and now unused) collection_move code.
For the record, the problem there is that a crash between setting the guard
and unlinking the old name will not remove the old name on replay because
the guard for the link stage is indistinguishable from that for the unlink
stage.
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
Sage Weil [Fri, 13 Apr 2012 16:56:04 +0000 (09:56 -0700)]
filestore: implement collection_move() as add + remove
This ensures we get add and remove steps with different spos values, which
makes the guard work. The collection_move implementation breaks on replay
because those values match, so the just-set guard prevents unlink from
happening.
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
ObjectStore: Remove code duplication when dumping transactions.
By using OStreamFormatter, we can have a single function responsible for
dumping a transaction. We keep the same old functions for outputting
directly to a Formatter and to an ostream, but these are only wrappers
for the function that will really handle the dumping.
The "real" dump() function will now take only a Formatter as an argument,
to which we will output. We keep the 'dump(std::ostream& out)' function,
although now it simply creates an OStreamFormatter and passes it to the
dump() function.
Signed-off-by: Joao Eduardo Luis <jecluis@gmail.com>
Sage Weil [Wed, 11 Apr 2012 19:33:57 +0000 (12:33 -0700)]
README: update
- refer to COPYING, SubmittingPatches
- a word about dependencies
- building packages
- drop the list of built binaries... it's incomplete and not very
interesting.
Sage Weil [Tue, 10 Apr 2012 22:24:49 +0000 (15:24 -0700)]
filestore: fix collection_move guard
We had a sequence like:
1- write A block 1
2- write A block 2
3- write A block 3
4- write A block 4
5- move A -> B
- link B
- unlink A
- set guard on B <crash>
- replay 3, 4, 5
with the result being B with only half of its content. The problem is that
we destroyed the old link _and_ didn't guard the new content. Instead,
set the guard before the link, and replay the unlink step here
unconditionally.
Fixes: #2178 Signed-off-by: Sage Weil <sage.weil@dreamhost.com>
projects / ceph.git / log