schedule, checker: add ruleChecker #2042
Conversation
Signed-off-by: disksing <i@disksing.com>
Signed-off-by: disksing <i@disksing.com>
Signed-off-by: disksing <i@disksing.com>
Signed-off-by: disksing <i@disksing.com>
Codecov Report
@@ Coverage Diff @@
## master #2042 +/- ##
==========================================
- Coverage 76.87% 76.84% -0.03%
==========================================
Files 184 185 +1
Lines 18653 18829 +176
==========================================
+ Hits 14339 14470 +131
- Misses 3227 3253 +26
- Partials 1087 1106 +19
Continue to review full report at Codecov.
|
| log.Warn("lost the store, maybe you are recovering the PD cluster", zap.Uint64("store-id", peer.StoreId)) | ||
| return false | ||
| } | ||
| return !store.IsUp() |
There was a problem hiding this comment.
Here we treat tomestone as offline too. (it happens when force bury a store before moing out all regions). replicaChecker's behaviour is like this too: https://github.com/pingcap/pd/blob/4d65bbefbc6db6e92fee33caa97415274512757a/server/schedule/checker/replica_checker.go#L216-L218
There was a problem hiding this comment.
Yes, but this function named isOfflinePeer which may be a little bit confused.
There was a problem hiding this comment.
em. do you have better suggestion? (replicaChecker uses checkOfflinePeer as function name too...)
There was a problem hiding this comment.
I think IsOfflineOrTomestone looks too verbose 😢
Signed-off-by: disksing <i@disksing.com> Co-Authored-By: Ryan Leung <rleungx@gmail.com>
Signed-off-by: disksing <i@disksing.com>
|
|
||
| // SelectStoreToReplacePeerByRule selects a store to replace a region peer in order to fit the placement rule. | ||
| func SelectStoreToReplacePeerByRule(scope string, cluster opt.Cluster, region *core.RegionInfo, fit *placement.RegionFit, rf *placement.RuleFit, peer *metapb.Peer, filters ...filter.Filter) *core.StoreInfo { | ||
| rf2 := removePeerFromRuleFit(rf, peer) |
There was a problem hiding this comment.
What purpose for this, don't want to remove looseMatchPeer?
There was a problem hiding this comment.
Similar to replicaChecker.SelectBestReplacementStore, replace is done by
copy region with the old peer removed + add a new peer to the copied region
The difference is in ruleChecker, replace happens inside a rule (remove a peer, add another peer fits the same rule), so the replace process besomes
copy RuleFit with the old peer removed + add a new peer to the copied RuleFit
Signed-off-by: disksing <i@disksing.com>
|
PTAL @rleungx |
rleungx
added
the
status/can-merge
|
/run-all-tests |
What problem does this PR solve?
Introduce
RuleCheckerto fix region according to placement rules.What is changed and how it works?
Check List
Tests