scheduler: new hotspot scheduler basic part #1870
Conversation
Luffbee
added
type/enhancement
|
/rebuild |
|
/rebuild |
|
PTAL @rleungx |
server/schedulers/hot_region.go
Outdated
| // the flow of the region exceeds the threshold, the scheduler will update the region in | ||
| // the hot cache and the hotdegree of the region will increase. | ||
| for storeID, items := range storeHotPeers { | ||
| storeHots, ok := stats[storeID] |
There was a problem hiding this comment.
storeHots sounds not good enough?
There was a problem hiding this comment.
What about hotRegions? Or hotPeers?
There was a problem hiding this comment.
Both are better than storeHots and I prefer to unify the naming of HotPeerStat and storeHotRegionsDefaultLen
There was a problem hiding this comment.
Unified similar names.
Codecov Report
@@ Coverage Diff @@
## master #1870 +/- ##
==========================================
- Coverage 78.03% 77.85% -0.18%
==========================================
Files 161 161
Lines 16054 16126 +72
==========================================
+ Hits 12527 12555 +28
- Misses 2542 2573 +31
- Partials 985 998 +13
Continue to review full report at Codecov.
|
| //| 4 | 6MB | | ||
| //| 5 | 0MB | | ||
| //| 6 | 0MB | | ||
| tc.UpdateStorageWrittenBytes(1, 7.5*MB*statistics.StoreHeartBeatReportInterval) |
There was a problem hiding this comment.
statistics.StoreHeartBeatReportInterval is too long.
There was a problem hiding this comment.
I noticed that, but I didn't find any better way to solve it.
Co-Authored-By: lhy1024 <admin@liudos.us>
Luffbee
added
the
status/can-merge
|
/run-all-tests |
|
@Luffbee merge failed. |
|
/merge |
|
/run-all-tests |
What problem does this PR solve?
Currently, the hotspot scheduler use the number of hot regions to make schedule decision, which cannot cover "many warm region" case, and may make wrong schedule decisions.
What is changed and how it works?
Use store flow byte rate to make decision.
Check List
Tests
Related changes