You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the following AWS dev test environment for test 1M nodes/ 5 regions /10 RPs per region/ 20K nodes per RP / 200k nodes per region / 40 schedulers / 25K nodes per scheduler, use PR100 "Split metrics report for new node events and update node events" to do the test.
test environment:
-- 1 service_api running on AWS EC2 instance (t2.2xlarge - 8 vcpu and 32GB memory)
-- 5 resource region simulators on 5 AWS EC2 instances in 5 different AWS regions(us-west-2, us-west-1, us-east-2, us-east-1 and ca-central-1)
* each EC2 instance simulates 200K nodes per region, 10RPs, 20K nodes per RP
-- 41 scheduler client simulators on 3 AWS EC2 instances in 2 different AWS regions (us-west-2 and us-west-1)
* 14 scheduler client simulators on 1 AWS EC2 instances in AWS region us-west-2
* 14 scheduler client simulators on 1 AWS EC2 instances in AWS region us-west-2
* 13 scheduler client simulators on 1 AWS EC2 instances in AWS region us-west-1
git log output:
commit 5e23651e28ebd9bdfdd5ee523881613c3de09f42 (HEAD, ying/perf-checkpoint)
Author: Ying Huang <[email protected]>
Date: Thu Jul 21 20:36:53 2022 +0000
Split metrics report for new node events and update node events
commit 526147a811c1704675cdde2f460f64804d78e33b (ying/main, origin/main, origin/HEAD, main, Carl_Test_By_PR93)
Author: Ying Huang <[email protected]>
Date: Thu Jul 21 09:01:52 2022 -0700
Change checkpoints from map to array (#93)
* Revert unnecessary changes in PR 85
* Move feature check out of business logic
* Fix spelling error
* Change checkpoints from map to array
* Add test case to TestSingleRPMutipleClients_Workflow: 1M nodes with 50 clients each has 15000 , each got 100K update events
* Print registration result properly
* Log latency detail for each event
* Add back testcases
* Use constants for checkpoint name
* Add perf data for distributor concurrency test after adding checkpoints with array
* Update per CR
commit 1ac3b09c60c1a25131f8ca6bbf718a9bf94c5730
Author: Yunwen Bai <[email protected]>
Date: Wed Jul 20 19:04:28 2022 -0700
aggregator PULL() optimization and logging adjust (#90)
* perf optimization and high logging level adjust
* minor fix to reduce client log
We only found 14 added events from service log /home/ubuntu/TMP/service.log.2022-07-21.v000060 under account 'ubuntu'.
did some ad hoc and calculation of watch node event numbers from simulator log and service metrics, the event numbers match to prove that there is no data change event lost.
In the following AWS dev test environment for test 1M nodes/ 5 regions /10 RPs per region/ 20K nodes per RP / 200k nodes per region / 40 schedulers / 25K nodes per scheduler, use PR100 "Split metrics report for new node events and update node events" to do the test.
test environment:
git log output:
We only found 14 added events from service log /home/ubuntu/TMP/service.log.2022-07-21.v000060 under account 'ubuntu'.
Initial investigation: Add node event were not in event queue. However, for watch related event (i.e. update) we won's lost them.
The text was updated successfully, but these errors were encountered: