Commit Graph

98 Commits

Author SHA1 Message Date
Meng Xu bd345f85db ConsistencyCheck:Fix failue due to address inconsistency between process and worker
With TLS, a worker (or process) can have a TLS address and non-TLS address.
When a process is created in simulation, the primary address is TLS by default.
The non-TLS one is the TLS address port plus one.

In a connection between two workers, if their primary addresses do not enable
or disable TLS together, one worker will swap its primary address and secondary address
so that the TLS config of the two endpoints can match.

The swap can make the primary address no longer the TLS one that was created
when the process is created. And the swap only happens for worker instead of
process struct in simulation.

This swap can cause worker->address != process->address.
In checkForExtraDataStores actor, we use worker->address to check if a process
is killable and use the process->address to kill the process. The inconsistency
can cause simulation to kill a protected process that is not killable and leads
to simulation failure.
2020-03-10 21:07:16 -07:00
Evan Tschannen 303df197cf Merge branch 'release-6.2'
# Conflicts:
#	CMakeLists.txt
#	bindings/c/test/mako/mako.c
#	documentation/sphinx/source/release-notes.rst
#	fdbbackup/backup.actor.cpp
#	fdbclient/NativeAPI.actor.cpp
#	fdbclient/NativeAPI.actor.h
#	fdbserver/DataDistributionQueue.actor.cpp
#	fdbserver/Knobs.cpp
#	fdbserver/Knobs.h
#	fdbserver/LogRouter.actor.cpp
#	fdbserver/SkipList.cpp
#	fdbserver/fdbserver.actor.cpp
#	flow/CMakeLists.txt
#	flow/Knobs.cpp
#	flow/Knobs.h
#	flow/flow.vcxproj
#	flow/flow.vcxproj.filters
#	versions.target
2020-03-06 18:22:46 -08:00
Evan Tschannen 1076abdee5 fixed crash when interf was not created 2020-03-05 19:09:08 -08:00
Evan Tschannen 1128666840 added additional logging on the log router 2020-03-05 18:17:06 -08:00
Evan Tschannen 96258b9809 Merge branch 'release-6.2'
# Conflicts:
#	documentation/sphinx/source/release-notes.rst
#	fdbcli/fdbcli.actor.cpp
#	fdbclient/ManagementAPI.actor.cpp
#	fdbrpc/FlowTransport.actor.cpp
#	fdbserver/ClusterController.actor.cpp
#	fdbserver/DataDistribution.actor.cpp
#	fdbserver/DataDistribution.actor.h
#	fdbserver/DataDistributionQueue.actor.cpp
#	fdbserver/KeyValueStoreMemory.actor.cpp
#	fdbserver/MasterProxyServer.actor.cpp
#	fdbserver/QuietDatabase.actor.cpp
#	fdbserver/SkipList.cpp
#	fdbserver/StorageMetrics.actor.h
#	fdbserver/TLogServer.actor.cpp
#	fdbserver/fdbserver.actor.cpp
#	fdbserver/storageserver.actor.cpp
#	fdbserver/workloads/KVStoreTest.actor.cpp
#	flow/CMakeLists.txt
#	flow/Knobs.cpp
#	flow/Knobs.h
#	flow/genericactors.actor.cpp
#	flow/serialize.h
2020-02-21 19:09:16 -08:00
Evan Tschannen cf4efca852 fix: buffered cursor should always make sure all of the sub-cursors are completely exhausted before calculating minVersion. It is not legal to advance a cursor version past an epochEnd (+100 million versions) without also returning the epochEnd mutation, or the storage servers might not be able to rollback far enough because the end of the previous epoch will be made durable 2020-02-19 15:24:32 -08:00
Alex Miller 7798456201 Make TLogs have consistent parallel peek behavior.
TLogServer and LogRouter had some leftover code from me trying to be
more "correct" about parallel peek semantics, but those changes weren't
reflected in the OldTLog* files.  I've reverted the changes, as
realistically, they are more likely to waste CPU than improve TLog behavior.
2020-01-21 18:23:16 -08:00
Alex Miller 858e4e5900 Move the check to a better location.
This way, we avoid some ID randomness, and also avoid the potential for
resetting the randomID and sequence without clearing out the future
vector.
2020-01-21 17:08:42 -08:00
Alex Miller 1cb311fcb8 Add an ASSERT_WE_THINK that peek cursors don't get timed_out()
This should prevent us from regressing and having multi-region
recoveries hang for 10min again.
2020-01-21 17:07:37 -08:00
Alex Miller 0662f8dba0 When switching parallel->single->parallel, reset sequence and peekId
This fixes an issue where one could hang for 10min for the second
parallel peek to time out, if one happened to catch the edge of a
onlySpilled transition wrong.
2020-01-21 17:07:37 -08:00
Evan Tschannen afc9713005 Merge branch 'release-6.2'
# Conflicts:
#	CMakeLists.txt
#	documentation/sphinx/source/release-notes.rst
#	fdbclient/FDBTypes.h
#	fdbserver/LogSystem.h
#	fdbserver/LogSystemPeekCursor.actor.cpp
#	fdbserver/OldTLogServer_6_0.actor.cpp
#	fdbserver/TLogServer.actor.cpp
#	versions.target
2019-11-06 13:45:37 -08:00
Evan Tschannen dbc5a2393c combineMessages still did not serialize tags correctly 2019-11-05 18:44:30 -08:00
Evan Tschannen 1c873591be fixed a compiler error 2019-11-05 18:32:15 -08:00
Evan Tschannen 86560fe727 fix: tempTags was not used correctly 2019-11-05 18:22:25 -08:00
Evan Tschannen a8ca47beff optimized memory allocations by using VectorRef<Tag> instead of std::vector<Tag> 2019-11-05 18:07:30 -08:00
Evan Tschannen daac8a2c22 Knobified a few variables 2019-11-04 20:21:38 -08:00
Evan Tschannen 457896b80d remote logs use bufferedCursor when peeking from log routers to improve performance
bufferedCursor performance has been improved
2019-11-04 19:47:45 -08:00
Evan Tschannen 3325980c03 Merge branch 'release-6.2'
# Conflicts:
#	CMakeLists.txt
#	documentation/sphinx/source/release-notes.rst
#	fdbserver/DataDistribution.actor.cpp
#	fdbserver/OldTLogServer_6_0.actor.cpp
#	fdbserver/TLogServer.actor.cpp
#	fdbserver/WorkerInterface.actor.h
#	fdbserver/worker.actor.cpp
#	versions.target
2019-10-24 17:38:15 -07:00
Evan Tschannen a7492aab0a fix: poppedVersion can update during a yield, so all work must be done immediately after getMore returns 2019-10-23 23:06:02 -07:00
Alex Miller 1e5b8c74e3 Continuing a parallel peek after a timeout would hang.
This is to guard against the case where

1. Peeks with sequence numbers 0-39 are submitted
2. A 15min pause happens, in which timeout removes the peek tracker data
3. Peeks with sequence numbers 40-59 are submitted, with the same peekId

The second round of peeks wouldn't have the data left that it's allowed
to start running peek 40 immediately, and thus would hang for 10min
until it gets cleaned up.

Also, guard against overflowing the sequence number.
2019-10-22 19:24:05 -07:00
Alex Miller c008e7f8b3 When switching parallel->single->parallel, reset sequence and peekId
This fixes an issue where one could hang for 10min for the second
parallel peek to time out, if one happened to catch the edge of a
onlySpilled transition wrong.
2019-10-22 19:10:58 -07:00
Evan Tschannen b495cc697b Merge branch 'release-6.2'
# Conflicts:
#	CMakeLists.txt
#	documentation/sphinx/source/release-notes.rst
#	versions.target
2019-09-13 09:25:08 -07:00
Alex Miller 324289039a When reloading one cursor in a merge cursor, top off the other cursors as well. 2019-09-12 16:22:28 -07:00
Jingyu Zhou 2723922f5f Replace -1 as VERSION_HEADER constant for serialization 2019-09-05 12:45:39 -07:00
Jingyu Zhou f9357c5ad8 Fix side effect of ArenaReader
ServerPeekCursor::nextMessage() should only consume the message header, because
the reader() directly inherits the current position. The previous commit
changes the positon to the begining of the next message, which breaks storage
server code.
2019-09-05 11:07:07 -07:00
Jingyu Zhou cd3f1e33d4 Refactor deserialization of TagsAndMessages
Consolidate deserialization of TagsAndMessages in the structure itself and
change both TLog and ServerPeekCursor to use it.
2019-09-04 14:55:05 -07:00
Evan Tschannen b0480edd15 fix: messageVersion could be larger than poppedVersion, and we will discard messages that are needed 2019-08-06 16:31:05 -07:00
Evan Tschannen 7ac7eb82f2 fix: buffered cursor would start multiple bufferedGetMore actors
advance all of the cursors to the poppedVersion
2019-07-30 14:42:05 -07:00
Evan Tschannen b5cb7919b6 fix: canDiscardPopped was not reset when necessary in all cases 2019-07-30 13:44:44 -07:00
Evan Tschannen 5d79e4141f fix: buffered cursor messageVersion should be set to the version we will be at after exhausting everything in messages 2019-07-30 12:38:44 -07:00
Evan Tschannen 6977e7d2e8 do not return recovered version as popped for txsTags because it could cause recovery to start over
optimized how buffered peek cursor discards popped data
2019-07-30 12:21:48 -07:00
Evan Tschannen 7a932479dd throw away state if we ever read popped data from the disk queue adapter 2019-07-30 10:14:39 -07:00
Evan Tschannen 45f7b41b48 fix: multi-cursor could discard popped commits after already returning data 2019-07-29 21:36:42 -07:00
Evan Tschannen 5bb322b483 implement popped on bufferedCursor 2019-07-29 21:19:47 -07:00
Evan Tschannen d8948c8be1 Merge branch 'master' into feature-fast-txs-recovery
# Conflicts:
#	fdbserver/TagPartitionedLogSystem.actor.cpp
2019-07-10 13:59:52 -07:00
Evan Tschannen b27a909f3a fix: onDisconnectOrFailure can spuriously trigger 2019-07-09 16:38:59 -07:00
Evan Tschannen 15e894c724 Merge in master 2019-07-05 15:49:24 -07:00
Evan Tschannen cfce1e1705 fix: buffered peek cursor would advance very slowly through large ranges of empty versions 2019-06-28 15:54:08 -07:00
Alex Miller bf883d7055 Merge remote-tracking branch 'upstream/master' into flowlock-api 2019-06-25 14:26:50 -07:00
Alex Miller 7a500cd37f A giant translation of TaskFooPriority -> TaskPriority::Foo
This is so that APIs that take priorities don't take ints, which are
common and easy to accidentally pass the wrong thing.
2019-06-25 02:47:35 -07:00
Evan Tschannen 1c005d5878
Merge pull request #1584 from alexmiller-apple/spilled-only-peek
Save TLog resources by letting peek request only spilled data.
2019-06-20 18:22:31 -07:00
Evan Tschannen e0be631414 shard the txs tag so that more transaction logs are involved in its recovery 2019-06-19 18:15:09 -07:00
Alex Miller ce24db3c53 Fully consume parallelPeekMore results before switching back. 2019-06-19 01:30:49 -07:00
Alex Miller 51fd42a4d2 Merge remote-tracking branch 'upstream/master' into spilled-only-peek 2019-06-18 17:33:52 -07:00
mpilman 8576665a90 Revert "Revert "Make protocol version a type""
This reverts commit 455bf3b3ec.
2019-06-18 14:49:04 -07:00
Alex Miller 455bf3b3ec Revert "Make protocol version a type" 2019-06-18 10:59:17 -07:00
mpilman da53a92bec Make protocol version a type
This fixes #1214

The basic idea is that ProtocolVersion is now its own type. This
alone is an improvement as it makes many things more typesafe. For
each version, we can now add breaking features (for example Fearless).
After that, there's no need to test against actual (confusing) version
numbers. Instead a developer can simply test
`protocolVersion->hasFearless()` and this will return true iff the
protocolVersion is newer than the newest version that didn't support
fearless.
2019-06-16 09:59:15 -07:00
Alex Miller 658e61b394 And now use spilledOnly as a hint to do parallel peeks.
If there's some spilled data, there's probably a lot of spilled data,
and now we can pull all of it faster.
2019-05-14 21:03:44 -10:00
Alex Miller 4eb4c03ce5 Save TLog resources by letting peek request only spilled data.
If a peek is entirely fulfilled from spilled data, then it's likely that
the next peek will be also.  It is thus wasteful for each of these peeks
to call peekMessagesFromMemory, which memcpy's excessively, and then
throw all that data away without using it.

Now, TLogs will give a hint back to peek cursors about if the provided
reply was served entirely from the spilled data, which peek curors then
feed back as the hint into their next request.

At some point, a cursor will send a request for only spilled data, get
an incomplete response, and then be told to send its next request as one
that peeks from memory as well, and then it will fully catch up.
2019-05-14 15:38:48 -10:00
A.J. Beamon 5f55f3f613 Replace g_random and g_nondeterministic_random with functions deterministicRandom() and nondeterministicRandom() that return thread_local random number generators. Delete g_debug_random and trace_random. Allow only deterministicRandom() to be seeded, and require it to be seeded from each thread on which it is used. 2019-05-10 14:01:52 -07:00