Closed ioga closed 3 months ago
Name | Link |
---|---|
Latest commit | 14ea8b016288be81168b1a92f3c6394896b5dbaa |
Latest deploy log | https://app.netlify.com/sites/determined-ui/deploys/6688327dfc4a0d00085a6392 |
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 52.88%. Comparing base (
0a57cde
) to head (14ea8b0
).
discussed offline, and we don't want to take the performance hit. we need a better way to generate the keys (in the background, reuse keys, ...) to alleviate the performance issues.
Description
It seems minimum recommended RSA key size on RHEL are 2048: https://access.redhat.com/articles/3642912 similar problem came up in the past for an openshift customer: https://hpe-aiatscale.atlassian.net/browse/DET-5983
I have a suspicion we're seeing it again for a fedora user: https://determined-community.slack.com/archives/CV3MTNZ6U/p1720179622244169
we elected not to do it in the past over performance concerns: https://github.com/determined-ai/determined/pull/2966#issuecomment-921397606 but now we
GenerateKey
once per experiment, not per trial.This'll slow down new experiment and shell creation by ~150 ms because the key generation takes longer.
Test Plan
det shell start
on a cluster with the default settingsopenssl rsa -text -in /run/determined/ssh/id_rsa | head -n1
Checklist
docs/release-notes/
See Release Note for details.