Use watch instead of list pods in node controller#1129
Use watch instead of list pods in node controller#1129centaurus-cloud-bot merged 4 commits intoCentaurusInfra:masterfrom
Conversation
…s instead of expensive listing pods call.
|
Can someone review this PR? |
|
can we test this change with a 1tp 2 rp cluster with smaller number of nodes? |
|
/lgtm. |
I tested in 2tp2rp 2x100 environment, perf test succeeded. However, we got unexpected error in KCM: This happened in 1tp1rp test as well. I will investigate. Otherwise, perf test works fine. |
The error seems happen only when cluster start up, not 100% occur, seems depends on the initialization time of the indexer and referral. And only once per node. Will add a log into POC branch to make sure it does not affect the actual index function. |
08d635c to
b036dc5
Compare
b036dc5 to
aa73519
Compare
Index was not added onto podinformer from tenant managers because informer was started before adding index. Rearranged informer start time, density test on 1x1x100 cluster. Index was properly added and expected debug log appeared. |
|
re-approve, the most recent scale test runs fine |
|
/approve |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: yb01, zmn223 The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Reduced # of list pods requests to ETCD. 1 request per 3 second, 1 call per node. This change significantly reduced system load during large cluster density test.