Before this PR, time elapse are logged after `compact()` finished
normally. No elapse will be logged with any errors.
This PR put elapse logs in defer once time-consuming steps are finished.
Signed-off-by: yangxuan <xuan.yang@zilliz.com>
- remove dl dependency while compiling some lib under MSYS
- int types aligement in golang
- core/src/utils adaptation for compile under MSYS
issue: #7706
Signed-off-by: Ji Bin <matrixji@live.com>
* Implement automatic item expiration on compaction
Signed-off-by: JunSung Park <writtic@gmail.com>
* Change entityExpiration value comparison logic
Signed-off-by: JunSung Park <writtic@gmail.com>
* Add logging for the number of expired entity
Signed-off-by: JunSung Park <writtic@gmail.com>
* Set default value of CompactionEntityExpiration to off
Signed-off-by: JunSung Park <writtic@gmail.com>
* Fix errors for code checker
Signed-off-by: JunSung Park <writtic@gmail.com>
* Set default entityExpiration to max value
Signed-off-by: JunSung Park <writtic@gmail.com>
When rootcoord drops a collection, it'll produce a `drop-collection DDL` msg
and one last timetick into DML channels of this collection.
So that when DataNode receives this msg, DN can release the resources for the specific collection.
Before this PR, RootCoord produced these two msgs with an older timestamp, generated
before many time-consuming RPCs.
Once these RPCs spend more time than timetick producing interval, the timetick of these 2 msgs are
older to the channel current timestamp, causing in-consistency in time, thus
making msgstream fail to consume the last `drop-collection DDL` msg.
This PR generates a new timestamp for `drop-collection DDL` msg and timetick msg after those time-consuming
RPCs
Fixes: #15406
Signed-off-by: yangxuan <xuan.yang@zilliz.com>
* Make SearchResults and QueryResults return collectionID
Signed-off-by: Filipe Caixeta <filipecaixeta@gmail.com>
* add return collection name from QueryRequest and SearchRequest
Signed-off-by: Filipe Caixeta <filipecaixeta@gmail.com>
* remove comment
Signed-off-by: Filipe Caixeta <filipecaixeta@gmail.com>
- Remove not useful log in `stopExecutingtaskByVChannelName`
- Add planID info in `mergeFlushedSegments`
Signed-off-by: yangxuan <xuan.yang@zilliz.com>
1. Add flowgraph manager to manager start and release of dataSyncService
2. Add event manager to manager etcd watchInfo event
- Make put event able to retry
- Make delete event able to terminate the retry loop
See also: #14604, #14300
Signed-off-by: yangxuan <xuan.yang@zilliz.com>
