Author: Kevin W Monroe
Author Date: 2024-03-24 23:37:40 UTC

deployment errors if all config isn't provided at deploy time (#5)

* lint/license adjustments

* spelling corrections, example usage

* consistent status messages, formatting, use block vs waiting appropriately

* no need for duplicate cni-joined and cni-changed relations; ensure state gates are bools; no update-status unless configured; need image secret before tigera, and license after

* fix non-leader premature kubectl calls

- non-leader units should wait for the cni relation before we consider
them 'configured' to ensure they'll have a valid kubeconfig available.
- waiting_for_cni in update_status is no longer needed since units won't
be 'configured' until that check passes.
- k-c-p may be in the middle of restarting the apiserver when we attempt
to query the tigera pod; handle that as if no pod is available. we'll
try again on the next hook that calls update_status.

* address failing integration tests

- test against stable bundle
- ensure model debug and halt on hook failure
- try/catch creation of install / crd manifests independently
- let our status handler fail since we're in a retry loop (fail well)

* retry kubectl calls (up to 3m) with tenacity

---------

Co-authored-by: Mateo Florido <32885896+mateoflorido@users.noreply.github.com>

Author: Kevin W Monroe
Author Date: 2024-03-24 23:37:40 UTC

deployment errors if all config isn't provided at deploy time (#5)

* lint/license adjustments

* spelling corrections, example usage

* consistent status messages, formatting, use block vs waiting appropriately

* no need for duplicate cni-joined and cni-changed relations; ensure state gates are bools; no update-status unless configured; need image secret before tigera, and license after

* fix non-leader premature kubectl calls

- non-leader units should wait for the cni relation before we consider
them 'configured' to ensure they'll have a valid kubeconfig available.
- waiting_for_cni in update_status is no longer needed since units won't
be 'configured' until that check passes.
- k-c-p may be in the middle of restarting the apiserver when we attempt
to query the tigera pod; handle that as if no pod is available. we'll
try again on the next hook that calls update_status.

* address failing integration tests

- test against stable bundle
- ensure model debug and halt on hook failure
- try/catch creation of install / crd manifests independently
- let our status handler fail since we're in a retry loop (fail well)

* retry kubectl calls (up to 3m) with tenacity

---------

Co-authored-by: Mateo Florido <32885896+mateoflorido@users.noreply.github.com>

Author: Kevin W Monroe
Author Date: 2024-03-24 23:37:40 UTC

deployment errors if all config isn't provided at deploy time (#5)

* lint/license adjustments

* spelling corrections, example usage

* consistent status messages, formatting, use block vs waiting appropriately

* no need for duplicate cni-joined and cni-changed relations; ensure state gates are bools; no update-status unless configured; need image secret before tigera, and license after

* fix non-leader premature kubectl calls

- non-leader units should wait for the cni relation before we consider
them 'configured' to ensure they'll have a valid kubeconfig available.
- waiting_for_cni in update_status is no longer needed since units won't
be 'configured' until that check passes.
- k-c-p may be in the middle of restarting the apiserver when we attempt
to query the tigera pod; handle that as if no pod is available. we'll
try again on the next hook that calls update_status.

* address failing integration tests

- test against stable bundle
- ensure model debug and halt on hook failure
- try/catch creation of install / crd manifests independently
- let our status handler fail since we're in a retry loop (fail well)

* retry kubectl calls (up to 3m) with tenacity

---------

Co-authored-by: Mateo Florido <32885896+mateoflorido@users.noreply.github.com>