Affects Version/s: 7.0.0, 7.1.0
Fix Version/s: 7.1.0
Running pyrock IG stress cloud test in GKE, on medium profile.
When initialising the test, pyrock run gatling against AM to populate CTS with tokens (session, etc...).
4 AM are adding tokens in parallel to 3 CTS instances, about 4000 tokens are added.
It is 100% reproducible with ig_access_token test on 7.1 and 7.0 releases
Gatling -> 4 AM -> 3 CTS
- Replication is ok with 2 CTS, but is broken with 3 CTS and 4 CTS as well.
- To avoid expiration tokens and deletion, i have set 1 day for token expiration and disabled TTL reaper on all CTS instances
- Looks like to not be a problem of resources, even with 12 cpu and 24 gig it does not work with 1000 tokens
The replication seems to be stabilised, the entry count is different but ds-sync-state is aligned on all instances.
When exporting ldif files and comparing the files, all files are not identical.
- Create a gke cluster
create static ip and cloud dns
connect to the cluster
change IP variable in cluster_create_step_2 script
- Checkout lodestar
- Checkout forgeops
- Configure Lodestar
- Add 1 day to expiration tokens in forgeops files
Create file (see maxSessionTime and maxSessionTime)
- Reduce number of users in pyrock test from 1 million to 1 000
- Add more resource to overseer
- Disable TTL reaper on all pods
- Run gatling