143f32b5bc
Update VTGateV3Features.md
...
Fixed review comments.
2015-11-30 17:51:44 -08:00
ea11f3ab5d
Rename VTGateV3.md to VTGateV3Features.md
2015-11-30 11:07:10 -08:00
013b5a85f0
Update VTGateV3.md
2015-11-28 00:19:23 -08:00
d6a68d3085
Update VTGateV3.md
2015-11-27 00:31:09 -08:00
bbfa773e98
Update VTGateV3.md
2015-11-26 22:17:54 -08:00
d2e9b11be6
Update VTGateV3.md
2015-11-26 21:38:04 -08:00
f969dc9e28
Create VTGateV3.md
2015-11-26 11:46:43 -08:00
a16cae4e14
Rename VTGateV3.md to VTGateV3Design.md
2015-11-26 11:25:18 -08:00
b8e3ae42e7
Merge pull request #1348 from michael-berlin/fix_cloud_storage_backup
...
GCS backup plugin: Adapt to new API of the GCS Go client.
2015-11-25 22:19:17 -08:00
aa9dfda4d5
GCS backup plugin: Adapt to new API of the GCS Go client.
...
This unbreaks our Go build.
In commit b467501b2a6e4998b369e3569864906a1261972c they changed their API from one-shot commands to a client.Bucket.Object pattern.
Where we previously stored only the "authCtx" in the global instance of the GSC backup plugin, we're now keeping the instance of the GSC Go client.
Note that we're supposed to call Close() on the GSC client instance eventually. However, our backup interface doesn't foresee a Close() yet. Therefore, for now, we keep the GSC Go Client instance around once it was actively used. The cost of this "leak" is therefore minimal.
NOTE: In their new API I didn't find a way to fill in the project id. We'll need to look into it.
NOTE: This code change is untested and may not work at all. However, it's good enough to merge and unbreak our build.
2015-11-25 18:22:23 -08:00
3de2db3272
GCS backup plugin: Fix typo.
2015-11-25 18:14:38 -08:00
2341bd4101
Merge pull request #1346 from youtube/suguwork
...
sqlparser: change JOIN syntax to match MySQL
2015-11-25 10:45:20 -08:00
b46ae81a95
v3 demo: fix to use modified vtaget API
2015-11-25 00:41:08 -08:00
4b059f1dcc
Merge pull request #1347 from enisoc/bench
...
Fix copy in KeyRangesOverlap
2015-11-23 23:43:04 -08:00
0dd6311ef6
sqlparser: Avoid unnecessary buffer pre-allocation.
...
bytes.Buffer has its own 64-byte bootstrap area, so there's no reason to
pre-allocate a buffer less than that.
Before:
```
BenchmarkParse1-12 20000 82960 ns/op 119243
B/op 101 allocs/op
BenchmarkParse2-12 10000 111208 ns/op 126882
B/op 315 allocs/op
```
After:
```
BenchmarkParse1-12 20000 71989 ns/op 119035
B/op 88 allocs/op
BenchmarkParse2-12 10000 111424 ns/op 126225
B/op 274 allocs/op
```
2015-11-23 18:47:18 -08:00
c8e300f0ea
vtgate: Clean up mapKeyRangesToShards()
...
Removed some unnecesary loops and allocations. Also there was no error
path in resolveKeyRangeToShards().
Before:
```
BenchmarkResolveKeyRangeToShards-12 5000000 326 ns/op
48 B/op 2 allocs/op
```
After:
```
BenchmarkResolveKeyRangeToShards-12 10000000 179 ns/op
0 B/op 0 allocs/op
```
2015-11-23 18:14:58 -08:00
f7bbee3f9a
Fix copy in KeyRangesOverlap.
...
The form &(*first) was being used, which doesn't actually make a copy.
Also got rid of some unnecessary allocations elsewhere in key.go.
Before:
```
BenchmarkUint64KeyBytes-12 1000000 1364 ns/op
720 B/op 15 allocs/op
BenchmarkUint64KeyString-12 1000000 1778 ns/op
800 B/op 20 allocs/op
BenchmarkKeyRangeContains-12 20000000 105 ns/op
0 B/op 0 allocs/op
BenchmarkKeyRangesIntersect-12 100000000 22.8 ns/op
0 B/op 0 allocs/op
BenchmarkKeyRangesOverlap-12 30000000 44.4 ns/op
0 B/op 0 allocs/op
```
After:
```
BenchmarkUint64KeyBytes-12 50000000 25.0 ns/op
0 B/op 0 allocs/op
BenchmarkUint64KeyString-12 10000000 121 ns/op
0 B/op 0 allocs/op
BenchmarkKeyRangeContains-12 20000000 69.8 ns/op
0 B/op 0 allocs/op
BenchmarkKeyRangesIntersect-12 100000000 14.3 ns/op
0 B/op 0 allocs/op
BenchmarkKeyRangesOverlap-12 20000000 105 ns/op
48 B/op 1 allocs/op
```
2015-11-23 18:10:58 -08:00
09d5960edb
sqlpraser: allow multiple tables when parenthesized
...
This something that MySQL allows.
2015-11-23 00:21:28 -08:00
8084edfbf2
sqlparser: change JOIN syntax to match MySQL
...
Specified here: http://dev.mysql.com/doc/refman/5.7/en/join.html
2015-11-23 00:04:12 -08:00
93540773c5
Merge pull request #1345 from youtube/travis_go15
...
Travis: Switch to Go 1.5. Fix that unit_test_race fails when a race is found.
2015-11-22 20:20:58 -08:00
a597e8b8c4
tools/unit_test_race.sh: Simplify if expression.
2015-11-22 19:45:30 -08:00
29a89f2183
tabletmanager: Fix data race in timeout test.
...
The timeout test effectively turned the synchronous RPC into an asynchronous: Due to the hit timeout, the test method finished while the server was still executing the RPC. In such a case, the defer of the test method and the RPC server code raced for the field "fakeRPCAgent.slow".
2015-11-22 17:41:04 -08:00
941ce6279e
Fix make unit_test_race such that it fails in case of errors.
...
Before this, the command always succeeded, even if races were found.
I'm grepping over the output of go test -race to check if there is race because there is no other way to distinguish between a found race and a flaky test.
2015-11-22 17:35:22 -08:00
522d248a74
Merge pull request #1344 from michael-berlin/vtworkerclient_separate_package_for_test
...
vtworkerclient: Move client test suite into its own package.
2015-11-22 16:43:30 -08:00
eba54ce272
vtworkerclient: Move client test suite into its own package.
...
Before this change production binaries, which included vtworkerclient, unnecessarily also included all dependencies (e.g. the go/zk package) of the test suite.
vtctlclient has a similar structure. There we had already a separate package only for the client test suite. I've added comments in its package as well to document why we have a separate package.
2015-11-21 21:15:36 -08:00
1bb5269a7e
vtworkerclient: Shorten interface name.
...
From "VtworkerClient" to "Client". This is still not in line with the Go recommendations (because vtworkerclient.Client still stutters), but better than before.
2015-11-21 21:10:49 -08:00
6f95608221
vtworkerclient: Reduce visibility of flag.
...
It doesn't have to be exported.
2015-11-21 21:08:38 -08:00
ee59cba47a
Merge pull request #1343 from youtube/suguwork
...
sqltypes: improve Result.Copy
2015-11-20 09:52:42 -08:00
7ae31aa871
protos: rest of the protos renamed to ...pb
2015-11-19 22:03:19 -08:00
d5f6685162
sqltypes: improve Result.Copy
...
Consolidate allocations to one per row, since we know that uses
less CPU.
2015-11-19 17:22:48 -08:00
15903a4d8a
Merge pull request #1342 from alainjobart/replication
...
Switching tabletconn API out of tproto.QueyrResultList.
2015-11-19 16:45:27 -08:00
aa39d1ed79
Merge pull request #1341 from alainjobart/resharding
...
Using ShardUpdateFields in reparent in a few places.
2015-11-19 16:45:16 -08:00
904509fb26
Handling no-update case better in UpdateShardFields.
...
(this was needed anyway in some other code).
2015-11-19 16:31:52 -08:00
3874c68ee9
Switching tabletconn API out of tproto.QueyrResultList.
2015-11-19 16:22:40 -08:00
e10162e139
Using ShardUpdateFields in reparent in a few places.
...
Will be more resilient to underlying shard updates.
2015-11-19 11:26:11 -08:00
b7743bdae1
Merge pull request #1340 from alainjobart/resharding
...
Removing a few obsolete BSON structures.
2015-11-19 09:20:05 -08:00
236e754965
Removing a few obsolete BSON structures.
2015-11-19 08:36:01 -08:00
1d1fd3e271
Merge pull request #1339 from alainjobart/resharding
...
One change needed for import.
2015-11-19 06:39:56 -08:00
527c9fe5b3
One change needed for import.
2015-11-18 20:18:03 -08:00
04c5b30d1f
Merge pull request #1337 from alainjobart/resharding
...
Refactoring vtgate service errors.
2015-11-18 17:29:06 -08:00
9362c57c61
Small fixes to vtgate unit tests to match empty arrays.
2015-11-18 17:11:40 -08:00
4dc2aabc9c
Merge branch 'master' into resharding
2015-11-18 16:59:57 -08:00
c7f000233d
Cosmetic changes to this test.
2015-11-18 16:59:27 -08:00
9e7778271a
Merge pull request #1338 from youtube/suguwork
...
sqltypes: backward compatibilty tweaks
2015-11-18 15:33:55 -08:00
d7a6e87f93
sqltypes: backward compatibilty tweaks
...
Some changes of sqltypes were not backward compatible:
- Sometimes we get nil bytes for valid string types.
- bsonrpc and grpc were different for Result.Rows.
2015-11-18 14:53:43 -08:00
49febf6b93
Merge pull request #1336 from enisoc/healthcheck
...
healthcheck: Don't update tablet record if it's been taken over.
2015-11-18 13:03:58 -08:00
cb85c6ab1f
Removing a couple more unused / useless methods.
2015-11-18 11:53:13 -08:00
74d0ec8441
Now renaming vtgate/proto to vtgate/gorpcvtgatecommon
2015-11-18 11:03:55 -08:00
28af24f2b7
Removing the last uses of vtgate/proto.
2015-11-18 10:53:48 -08:00
309687b3e5
Changing vtgate's Begin call to return the session.
2015-11-18 10:04:31 -08:00
sougou
Michael Berlin
Anthony Yeh
Alain Jobart
Alain Jobart