ambassador v1 api #492
ambassador v1 api #492
Conversation
|
Actually when I run |
|
With 561157c it now passes for me. I was able to recreate the behaviour in a notebook, though it does not happen every time so was not easy to reproduce. Seems like it's not just that the first grpc request after deployment fails, it's that sometimes it can take minutes for the grpc to start working, even after rollout reports completion (which is longer than the retries were allowing for) |
|
After discussion on the ambassador slack and the linked ambassador issue, have found we can get the grpc requests working in a timely fashion by running ambassador as root so have updated PR to do that |
|
Turns out grpc requests are better but still intermittent problems - #498 |
* Fix strimzi helm values bug for zookeeper persistent storage * fix kafka bootstrap location to be internal SVC
* fix small typo * do not set default min replicas * consider model loaded on replica when filtering * add unit test * add more tests * add very simple crude model scaling logic * refactor function * lint fixes * lint fixes * add test * reorder funcs * add smoke test * fix func name * add scaling down test * refactor * lint * simplify test * set defaults if min / max replicas are missing * add range checks * tidy up scaling range logic * more tidy up * add more tests * add cooling off period * fix lint * add range checks for replicas * remove unused env variables * remove lock * revert REQUESTS_CA_BUNDLE (#491) * tidy up notebook with more models for triton (#494) * Fix Strimzi Helm values ZK indent bug + stale broker service name (#492) * Fix strimzi helm values bug for zookeeper persistent storage * fix kafka bootstrap location to be internal SVC * fix log message * add env variables to runner * tidy up comment * switch log to debug * do not update k8s meta if nil * tidy up ,model name * fix decrement bug Co-authored-by: cliveseldon <[email protected]>
fixes #491
I checked against the specs referenced in that issue and did a step through of the timeouts notebook to check that feature works.
I also ran the E2E tests - I did get a failure but it looks like an instance of #473 Surprised to see that even with the retry library there though