Commit Graph

418 Commits (d512b5f98b7133ccd0c0261570e31e78c0ff484a)

Author SHA1 Message Date
yah01 94a5122044
Fix LoadBalance removes source nodes from DmChannel (#17744)
Only NodeDown LoadBalance should remove source nodes

Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-24 12:40:14 +08:00
wayblink 1a336bbbf1
Fix: WatchDmChannelsRequest can be too large to save in etcd (#17722)
Signed-off-by: wayblink <anyang.wang@zilliz.com>
2022-06-24 10:54:15 +08:00
yah01 b4f21259ef
Fix SegmentChangeInfo is not set properly (#17701)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-22 21:08:13 +08:00
yah01 a077bad84a
Make updating replica atomic, balance plan idempotent (#17686)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-22 13:40:13 +08:00
yah01 1e87b54542
Add the offlines nodes previously crashed to offline nodes channel (#17663)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-21 16:08:13 +08:00
wayblink f2bd910df5
[skip e2e]Fix log mistake: WatchDmChannels -> WatchDeltaChannels (#17643)
Signed-off-by: wayblink <anyang.wang@zilliz.com>
2022-06-20 15:10:12 +08:00
congqixia 50cecc65ed
Fix mock querynode server session not revoked (#17229)
Revoke mock querynode server session when it's stopped
This PR reduces the running time of TestLoadBalanceIndexedSegmentsAfterNodeDown from 60+ seconds to less than 1+ seconds
Also related to #17212 #17215

Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-06-18 18:22:12 +08:00
wayblink 074ec3060a
Support return dropped segments info in GetSegmentInfo rpc (#17617)
Signed-off-by: wayblink <anyang.wang@zilliz.com>
2022-06-17 18:24:12 +08:00
congqixia 785a5a757f
Use segment version instead of ref cnt (#17609)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-06-17 17:38:12 +08:00
congqixia cc3ecc4bd5
Make querycoord channel allocator respect context (#17552)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-06-17 16:02:12 +08:00
yah01 f5fa93aa0b
Return err if failed to assign segments/channels to nodes (#17616)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-17 12:42:11 +08:00
yah01 0f87763682
Fix LoadBalance may not update the shard leader (#17608)
This happens probably with concurrent updating replicas,
some goroutines modify the nodes list of replicas,
and the others modify the shard leaders of replicas

Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-17 08:58:10 +08:00
yah01 3f42f5f345
Set the task state to TaskFailed if error occurs (#17598)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-16 19:18:11 +08:00
Xiaofan 1f6fbf91b2
Fix pulsar unsubsribe issue (#17562)
Signed-off-by: xiaofan-luan <xiaofan.luan@zilliz.com>
2022-06-16 17:28:11 +08:00
yah01 7d5c8c5f38
Fix bug not remove offline node (#17560)
The LoadBalance task won't remove the offline node if the node never load/watch any segment/dmchannel

Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-16 16:00:10 +08:00
wayblink eb5b0b7fc8
Move SegmentInfo out of VchannelInfo, leave Id only to decrease kv size. Get complete SegmentInfo through RPC (#17441)
Resolves: #17233 #16047

Signed-off-by: wayblink <anyang.wang@zilliz.com>
2022-06-16 12:00:10 +08:00
cai.zhang ea5041aec2
Acquiring the segment reference lock on task level (#17544)
Signed-off-by: Cai.Zhang <cai.zhang@zilliz.com>
2022-06-15 21:38:10 +08:00
congqixia f9553970f9
Add BindContext function for querycoord task scheduler (#17531)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-06-14 17:12:09 +08:00
Enwei Jiao a5b008acec
ignore getReplica's error when handle rebalanceTask (#17469)
Signed-off-by: Enwei Jiao <jiaoew2011@gmail.com>
2022-06-10 14:50:08 +08:00
Xiaofan 66f26943f8
Add Err code when task rollback (#17472)
Signed-off-by: xiaofan-luan <xiaofan.luan@zilliz.com>
2022-06-09 21:08:07 +08:00
congqixia 97a871cc82
Make querycoord segment allocator respect context (#17452)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-06-09 20:14:06 +08:00
Letian Jiang dfaed5acdd
Add QueryServiceAvailable field in ShowCollections (#17456)
Signed-off-by: Letian Jiang <letian.jiang@zilliz.com>
2022-06-09 18:20:07 +08:00
yah01 a2d2ad88bd
Make assigning segments faster (#17377)
This improve the Load performance,
and let the LoadBalance fails fast, which allows us to retry it timely

Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-06 16:52:05 +08:00
xige-16 8c69790383
Fix lost delete msg caused by loadSegment after watchDeltaChannel (#17308)
Signed-off-by: xige-16 <xi.ge@zilliz.com>
2022-06-02 16:56:04 +08:00
yah01 cc69c5cdd3
Make Cluster interface's methods called outside public (#17315)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-02 13:16:05 +08:00
yah01 f5bd519e49
Add retry mechanism for NodeDown LoadBalance (#17306)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-06-01 20:00:03 +08:00
cai.zhang bcf3b7426a
Add distributed lock for segment refgerence (#16782)
Signed-off-by: Cai.Zhang <cai.zhang@zilliz.com>
2022-05-31 16:36:03 +08:00
congqixia c88514bc49
Remove not used QueryChannel in Proxy and Query Cluster (#16856)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-30 19:50:04 +08:00
yah01 b09359b12f
Remove useless collection ID in error message (#17269)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-30 14:40:01 +08:00
xige-16 3a63d6c98e
Fix load timeout in chaos (#17241)
Signed-off-by: xige-16 <xi.ge@zilliz.com>
2022-05-26 22:28:03 +08:00
Letian Jiang f2a27e0e64
Retry GetShardLeaders until service available or timeout (#17183)
Signed-off-by: Letian Jiang <letian.jiang@zilliz.com>
2022-05-26 20:28:02 +08:00
yah01 5872c5afb6
Fix updating shard leaders may lost some modifications (#17218)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-26 17:14:02 +08:00
congqixia 7409bfc56d
Make allocateNode run async in case of block offline event (#17185)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-25 18:53:59 +08:00
yah01 de0ba6d495
Fix GetQuerySegmentInfo() returns incorrect result after LoadBalance (#17190)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-25 15:17:59 +08:00
congqixia 37d7d7baf8
Fix node state data race in querycoord (#17198)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-25 10:43:59 +08:00
Jiquan Long 75ca64f8c7
Refine task type logs to string (#17196)
Signed-off-by: longjiquan <jiquan.long@zilliz.com>
2022-05-25 03:08:00 +08:00
Bingyi Sun ffaead6ad9
Add load meta in hand off task. (#17179)
issue: #16842

Signed-off-by: sunby <bingyi.sun@zilliz.com>

Co-authored-by: sunby <bingyi.sun@zilliz.com>
2022-05-24 18:24:00 +08:00
Bingyi Sun 86728490a2
Fix partition not found (#17132)
issue: #16842

Signed-off-by: sunby <bingyi.sun@zilliz.com>

Co-authored-by: sunby <bingyi.sun@zilliz.com>
2022-05-20 19:37:57 +08:00
yah01 7746a5b742
Add NodeIds field for QuerySegmentInfo (#17121)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-20 18:03:58 +08:00
congqixia 599763d9bf
Fix replicas info is not removed after release (#17111)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-19 20:05:57 +08:00
yah01 dcfe472586
Fix LoadBalance doesn't save the modification to replicas' shards (#17064)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-19 16:51:57 +08:00
bigsheeper 9eeec4a2d5
Add collection load cache and InvalidateCollMetaCache by collID (#16882)
Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
2022-05-19 10:13:56 +08:00
yah01 33c855dcd2
Fix LoadBalance doesn't remove the source nodes from segment (#17051)
If the triggerCondition isn't NodeDown, the removing won't happen.

Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-18 11:55:56 +08:00
yah01 960d35e517
Add MockCluster, make unit tests reliable (#17032)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-17 12:57:56 +08:00
Bingyi Sun 59bc0a7000
Add some log. (#16978)
Signed-off-by: sunby <bingyi.sun@zilliz.com>

Co-authored-by: sunby <bingyi.sun@zilliz.com>
2022-05-16 17:15:55 +08:00
yah01 e38c6f6c44
Fix load the same segments multiple times for manual LoadBalance (#16921)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-16 15:41:56 +08:00
yah01 a382133a8a
Add new node into the replica which has the most of offline nodes (#16907)
Signed-off-by: yah01 <yang.cen@zilliz.com>
2022-05-13 18:31:54 +08:00
congqixia ae717bf991
Fix channelUnsubscribe data race and logic (#16946)
- Add a RWMutex for container/list which is not goroutine-safe
- Fix the element in list is never removed

Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-12 18:09:53 +08:00
congqixia 5c98329f7c
Return error when no replica available (#16886)
Signed-off-by: Congqi Xia <congqi.xia@zilliz.com>
2022-05-10 19:47:53 +08:00
Xiaofan 000c5ff3de
Fix msgstream unsubscription (#16883)
Signed-off-by: xiaofan-luan <xiaofan.luan@zilliz.com>
2022-05-10 19:43:52 +08:00