bd62b0a646
mvcc: attach keys to leases after recover all state
...
The previous logic is wrong. When we have hisotry like Put(foo, bar, lease1),
and Put(foo, bar, lease2), we will end up with attaching foo to two leases 1 and
2. Similar things can happen for deattach by clearing the lease of a key.
Now we try to fix this by starting to attach leases at the end of the recovery.
We use a map to keep the last lease attachment state.
2016-08-04 11:17:58 -07:00
982e18d80b
*: regenerate proto with latest grpc-gateway
2016-07-27 13:21:03 -07:00
fffa484a9f
*: regenerate proto for adding deleterange
2016-07-23 16:17:44 -07:00
50be793f09
*: regenerate proto
2016-07-18 09:33:32 -07:00
ba2725c2d0
build, backend: add backend commit failpoints
2016-07-14 12:26:35 -07:00
c853704ac9
*: support get-old-kv in watch
2016-07-05 16:17:09 -07:00
bc6d7659af
Merge pull request #5795 from xiang90/filter
...
*: support watch with filters
2016-06-28 14:07:12 -07:00
dced92f8bd
*: support watch with filters
...
Now user can filter events with types. The API is also extensible.
It might make sense for the proxy to filter out events based on
more expensive/customized filter.
2016-06-28 13:46:57 -07:00
38035c8c13
Merge pull request #5794 from xiang90/fix_c
...
mvcc: do not hash consistent index
2016-06-28 12:25:32 -07:00
ef9754910e
mvcc: do not hash consistent index
2016-06-28 09:36:26 -07:00
9614dc6e71
etcdserver: check index of the kv when restarting
2016-06-27 10:27:27 -07:00
dbc7c2cf4e
backend: reuse timer in run().
...
Benchmarks:
```
import (
"testing"
"time"
)
func BenchmarkTimeAfter(b *testing.B) {
b.ReportAllocs()
for n := 0; n < b.N; n++ {
select {
case <- time.After(1 * time.Millisecond):
}
}
}
func BenchmarkTimerReset(b *testing.B) {
b.ReportAllocs()
t := time.NewTimer(1 * time.Millisecond)
for n := 0; n < b.N; n++ {
select {
case <- t.C:
}
t.Reset(1 * time.Millisecond)
}
}
```
Running reveals that each loop results in 3 allocs:
```
BenchmarkTimeAfter-4 2000 1112134 ns/op 192 B/op 3 allocs/op
BenchmarkTimerReset-4 2000 1109774 ns/op 0 B/op 0 allocs/op
```
2016-06-23 18:49:41 +05:00
def21f11a9
*: support count in range query
2016-06-21 16:20:55 -07:00
6f28b43806
*: fix pending events metrics
2016-06-19 23:00:39 -07:00
5a7b7f7595
main: add grpc-gateway support
...
Now etcd can serve HTTP json request at /v3alpha/
2016-06-14 17:09:06 -07:00
6bbd8b7efb
mvcc: add keyIndex benchmark test
...
Useful later when trying to optimize our restore operations.
2016-06-09 14:13:18 -07:00
1610391449
*: following changes for proto update
2016-06-07 13:33:03 -07:00
f57b4eb46d
mvcc: don't cancel watcher if stream is already closed
...
Close() already cancels all the watchers but doesn't bother to clear out
the bookkeeping maps so Cancel() may try to cancel twice.
Fixes #5533
2016-06-03 11:12:46 -07:00
9c767cbf98
Merge pull request #5464 from heyitsanthony/fix-victim-watchers
...
mvcc: tighten up watcher cancelation and revision handling
2016-05-30 20:09:39 -06:00
3266c809e4
mvcc: delete EXPIRE event type
...
Addressing https://github.com/coreos/etcd/pull/5484#discussion_r65005236 .
etcd v3 doesn't expire keys. It's either PUT of DELETE.
2016-05-29 14:54:38 -07:00
cfb3f96c2b
mvcc: tighten up watcher cancelation and revision handling
...
Makes w.cur into w.minrev, the minimum revision for the next update, and
retries cancelation if the watcher isn't found (because it's being processed
by moveVictims).
Fixes : #5459
2016-05-27 17:19:32 -07:00
bf8cf39daf
mvcc: use capnslog
2016-05-20 22:31:22 -07:00
394ce5f3b8
mvcc: move blocked unsynced watchers to victim list
2016-05-20 15:56:02 -07:00
5984e46364
mvcc: move blocked sync watcher work to victim list
...
Instead of holding the store lock while doing a lot of work like when syncung
unsynced watchers, the work from a blocked synced notify can be reused and
dispatched without holding the store lock for long.
2016-05-20 15:56:02 -07:00
0b34b236d6
mvcc: benchmark for synced watchers
2016-05-19 23:31:27 -07:00
aa85cf037f
mvcc: remove defer in watchable store
2016-05-19 13:51:51 -07:00
6f2e7875aa
etcdctl: add migrate command
...
Migrate command accepts a datadir and an optional user-provided
transformer function that transform v2 keys to v2 keys.
Migrate command then builds a v3 backend state based on the existing
v2 keys and the output of the transformer function.
2016-05-19 12:17:15 -07:00
77775e8e92
mvcc: preallocate bytes buffer for saveIndex
2016-05-18 10:01:57 -07:00
e4e4c9dc2c
mvcc: set bolt options to nil for non-linux systems
2016-05-17 12:46:44 -07:00
9d9f02c1ee
mvcc: update LICENSE header
2016-05-12 20:50:33 -07:00
080272be17
mvcc: limit total watchers synced per sync
...
Fixes #4567
2016-05-11 11:16:43 -07:00
0fb7cb8b00
*: add disk operation metrics for monitoring
2016-05-11 09:36:45 -07:00
34b0736f2c
mvcc: Reduce number of allocs when watchableStore if no watchers.
...
When there are no watchers the number of allocations made while handling
a PUT operation can be reduced by exiting early.
2016-05-11 00:51:00 -07:00
a288188001
*: typo, remove string type assertions
2016-05-03 10:59:57 -07:00
015acabdbb
*: rerun genproto -g
2016-05-02 23:02:31 -07:00
7a6d9ea01a
mvcc: Hash to return Revision
2016-05-02 15:04:24 -07:00
ad327e01d0
mvcc: add benchmark for watch put and improve it
2016-04-29 19:58:37 -07:00
3ddcc21179
mvcc: fix watch deleteRange
2016-04-29 09:40:28 -07:00
67645095e9
*: add debugging metrics
2016-04-26 09:52:56 -07:00
b7ac758969
*: rename storage package to mvcc
2016-04-25 15:25:51 -07:00
