1a4ca05a04
Merge pull request #1889 from ajbeamon/add-cache-memory-parameter
...
Add cache_memory parameter to fdbserver
2019-07-26 13:34:24 -07:00
98c3b24036
Merge pull request #1869 from alexmiller-apple/sharded-txs-performance
...
Raise the priority of TLogRejoin above TLogPeekReply
2019-07-26 13:30:13 -07:00
fc7f9e7c69
Fix Void flatbuffers serialization
...
Before this fix, the encoding of Void happens to 'work' in that this
implementation can both read and write it, but it is not a flatbuffers
message since it contains a invalid offset (the implementation 'works'
because it does not dereference that offset).
2019-07-25 17:30:43 -07:00
a92b6cd3d1
Merge branch 'master' into add-cache-memory-parameter
...
# Conflicts:
# documentation/sphinx/source/release-notes.rst
2019-07-25 13:41:57 -07:00
a65c9f92ed
get rid of all timeouts and other changes
2019-07-24 15:36:28 -07:00
a2f2ad96ff
code review comments and merge to master changes
2019-07-24 15:36:28 -07:00
806a0822f1
timeoutError that takes Future<Error<T>> as input
2019-07-24 15:36:28 -07:00
94be9560ea
Add cache_memory parameter to fdbserver to control the size of the (4K) page cache. Change the default slighty from 2000 MiB to 2GiB.
2019-07-23 15:05:21 -07:00
6e078a41a7
Merge pull request #1828 from mpilman/features/lexicographical-ordered-traces
...
Make trace files lexicographically ordered
2019-07-23 08:31:59 -07:00
e98cee016d
Fix unsafe usage of now() function from multiple threads in trace logging.
2019-07-22 22:31:38 -07:00
5cef65b6c4
Update flow/FileTraceLogWriter.cpp
...
Co-Authored-By: A.J. Beamon <ajbeamon@users.noreply.github.com>
2019-07-22 16:38:18 -07:00
de4fe4d640
Remove dead code and revert back to base 10
...
there can be at most 2^32 files with a distinct index. Therefore
`log10(index) < 10` - this means base 10 is good enough.
2019-07-22 16:20:03 -07:00
aa93f8411a
Merge branch 'features/lexicographical-ordered-traces' of github.com:mpilman/foundationdb into features/lexicographical-ordered-traces
2019-07-22 14:22:51 -07:00
001bfaf80c
made trace index unsigned
2019-07-22 14:20:09 -07:00
a48bdc0095
Update flow/FileTraceLogWriter.cpp
...
Co-Authored-By: A.J. Beamon <ajbeamon@users.noreply.github.com>
2019-07-22 14:16:34 -07:00
e29a6ea280
Merge pull request #1871 from bnamasivayam/tr-priority-add-client-log
...
Track the priority of sampled Transaction as part of GetReadVersion e…
2019-07-22 13:04:52 -07:00
c26c68d6d2
Address review comments
2019-07-22 11:45:05 -07:00
a66dc937fc
Make trace files lexicographically ordered
...
This resolves #1825
The basic idea is that we prefix every sequence number with its with,
separated by a dot. That way the filename length isn't increased
drastically (unlike using a fixed whidth number with 0-padding)
and the file names will be lexicographically ordered.
2019-07-22 10:10:47 -07:00
af267ba053
Track the priority of sampled Transaction as part of GetReadVersion event.
2019-07-19 17:31:49 -07:00
df7f0cffa1
Raise the priority of TLogRejoin above the default work priority.
...
With sharded txs tags, the master now receives data from transaction
logs at an order of magnitude higher rate. This is the intentional
desires result of sharding the txs tag. With a sufficient number of
TLogs, the master will saturate its CPU time handling the peek
responses.
Performance tests revealed some unstable oddities in how long a recovery
would take, which was eventually root caused to a priority inversion
between TLogRejoin requests and TLog peek replies.
Once peek replies saturate the CPU, the master would proceed to ignore
further TLogRejoin messages. TLogRejoin is what marks a TLog as
available to the failure monitor, which is also what decides between a
ServerPeekCursor and a MergePeekCursor for a SetPeekCursor. Ignoring
TLogRejoins meant that the sharded txs locality tags for those servers
would be merge peeked over all TLogs. This is much less efficient than
just peeking one copy of data from the one preferred server.
Depending on the race between TLogPeek replies saturating the CPU and
TLogRejoin requests being submitted, a variable number of tags would be
affected, and thus the performance test would have some variance in its
results.
2019-07-19 16:55:04 -07:00
3045826e3c
Merge pull request #1819 from mpilman/flatbuffers-fixes2
...
Flatbuffers fixes2
2019-07-19 16:33:50 -07:00
c3a8ae4752
Merge pull request #1791 from fzhjon/fetch-keys-requests-priority
...
Introduce priority to fetchKeys requests from data distribution
2019-07-19 14:54:51 -07:00
8e9aaa767e
Merge pull request #1860 from alexmiller-apple/comma-operator
...
Remove `operator , (vector<T>, T)` as an append operator.
2019-07-19 12:53:32 -07:00
9f0a9da7c3
Remove `operator , (vector<T>, T)` as an append operator.
...
Because no one expects the comma overload.
2019-07-18 21:16:08 -07:00
1ac2d01b03
Merge remote-tracking branch 'upstream/master' into flatbuffers-fixes2
2019-07-18 09:50:08 -07:00
b9e938972c
Merge pull request #1719 from mpilman/features/static-libc++
...
statically link libstdc++ on Linux and remove std::variant
2019-07-16 21:00:02 -07:00
5d3e69b6fc
Merge pull request #1820 from fzhjon/load-balance-locality
...
Introduced a knob that can turn locality on/off
2019-07-16 16:40:43 -07:00
6a4a129cf5
fixed silly boost visitor bug
2019-07-16 15:10:55 -07:00
b18666d942
statically link libstdc++ on Linux and remove std::variant
...
this will hopefully fix #1610
2019-07-16 14:53:16 -07:00
d5caf0c1b4
Merge branch 'flatbuffers-fixes2' of github.com:mpilman/foundationdb into flatbuffers-fixes2
2019-07-16 14:47:40 -07:00
e7e48a40ce
Fix a few IncludeVersion bugs
2019-07-16 13:28:29 -07:00
7d37040725
split the single knob into two for finer-grained control
2019-07-16 12:46:02 -07:00
a89d09f1a6
Merge pull request #1844 from ajbeamon/configurable-trace-field-length
...
Make trace event field lengths default knobified and individually configurable.
2019-07-15 15:53:22 -07:00
54416f46fd
Pass type as param to VectorRef instead of bool
2019-07-15 15:08:49 -07:00
562e41d25d
Apply suggestions from code review
...
Co-Authored-By: Alec Grieser <alloc@apple.com>
2019-07-15 15:02:55 -07:00
707b29cbcc
Fixed compilation error
2019-07-15 13:33:22 -07:00
c8e94e601a
Merge pull request #1729 from etschannen/feature-fast-txs-recovery
...
Improve the recovery speed of the txnStateStore
2019-07-15 13:27:41 -07:00
6c6a1ca8f4
Expose serialization context too all traits
2019-07-15 12:58:31 -07:00
1ab36cafbc
Fixed memory bug
2019-07-15 09:53:26 -07:00
cb7b941452
expose protocolVersion to serialize function
...
The main idea here is that we now have a context for
serialization and deserialization. This can be used
to access the protocolVersion.
The context still needs to be exposed to the
deserailization traits
2019-07-12 17:16:54 -07:00
d5051b08dd
Make trace event field lengths (and total event sizes) default knobified and configurable. Add a transaction option to control the field length of transaction debug logging. Make the program start command line field less likely to be truncated.
2019-07-12 16:12:35 -07:00
75d4b612cf
Make object serializer versioned
2019-07-12 11:53:14 -07:00
984b603c33
Avoid saving each item twice
2019-07-12 11:20:01 -07:00
11c969511c
Update TODOs
2019-07-12 11:17:11 -07:00
74f992b2cc
fixed wrong assertion
2019-07-11 23:03:31 -07:00
b68f2d925f
Serialize range result to string for speed
2019-07-11 23:03:31 -07:00
c63ff98535
Add some documentation todos
2019-07-11 23:03:31 -07:00
eb87b1fc61
Remove dead code
2019-07-11 23:03:31 -07:00
137936762d
Eliminate unnecessary memcpy
2019-07-11 23:03:31 -07:00
70f0726185
Support PacketBuffer's of arbitrary size
2019-07-11 23:03:31 -07:00
Evan Tschannen