Add PodDisruptionBudgets to the Pinot Helm chart

[andimiller]

I've set these by default at allowing 50% of the servers/brokers/controllers to be unavailable, and exposed these settings in the configuration so the user can change them.

These prevent outages when the underlying kuberenetes cluster undergoes an upgrade.

You might configure these like so in your helm yaml settings:

broker:
  replicaCount: 2
  pdb:
    enabled: true
    minAvailable: 50% # we could also set this to 1
    maxUnavailable: "" # can turn these off with empty string
controller:
  replicaCount: 2
  pdb:
    enabled: true
    minAvailable: 1 # we could also set this to 50%
    maxUnavailable: "" # can turn these off with empty string
server:
  replicaCount: 12
  pdb:
    enabled: true
    maxUnavailable: 1
    # minAvailable: 11 # these would be equivalent

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 62.25%. Comparing base (59551e4) to head (b2dfe14).
Report is 470 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff              @@
##             master   #13153      +/-   ##
============================================
+ Coverage     61.75%   62.25%   +0.50%     
+ Complexity      207      198       -9     
============================================
  Files          2436     2529      +93     
  Lines        133233   138215    +4982     
  Branches      20636    21386     +750     
============================================
+ Hits          82274    86048    +3774     
- Misses        44911    45762     +851     
- Partials       6048     6405     +357     

Flag	Coverage Δ
custom-integration1	<0.01% <ø> (-0.01%)	⬇️
integration	<0.01% <ø> (-0.01%)	⬇️
integration1	<0.01% <ø> (-0.01%)	⬇️
integration2	0.00% <ø> (ø)
java-11	62.19% <ø> (+0.48%)	⬆️
java-21	62.14% <ø> (+0.51%)	⬆️
skip-bytebuffers-false	62.22% <ø> (+0.47%)	⬆️
skip-bytebuffers-true	62.12% <ø> (+34.39%)	⬆️
temurin	62.25% <ø> (+0.50%)	⬆️
unittests	62.25% <ø> (+0.50%)	⬆️
unittests1	46.75% <ø> (-0.14%)	⬇️
unittests2	27.92% <ø> (+0.18%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[abhioncbr]

Thanks for the contribution.

In general, one comment from K8s before v1.21 support, and to give other users better control over the PDB, can we have a boolean property say enabled to make the PDB optional?

[andimiller]

Thanks for the contribution.

In general, one comment from K8s before v1.21 support, and to give other users better control over the PDB, can we have a boolean property say enabled to make the PDB optional?

Sure, done, and since the default replica counts of everything in values.yaml is 1 I've defaulted it to off

[Jackie-Jiang]

cc @xiangfu0 @zhtaoxiang

[gortiz]

FYI, the zookeeper chart from bitnami (the one we depend on) also has this ability.

[gortiz]

It would be nice to also have the ability to define a maxUnavailable.

This is specifically useful in servers, where if we know the min replica size for all tables, we can configure pdb as maxUnavailable: replicaSize - 1

[andimiller]

added

[andimiller]

I've left the defaults values for broker and controller at 50%, although replicaSize - 1 would make sense too

The number for servers would be maxUnavailable: $tableReplicas - 1 or similar, which isn't easily available here so I just set it to 1 in the default values

[andimiller]

FYI, the zookeeper chart from bitnami (the one we depend on) also has this ability.

I'll adjust this to work with the same config settings as the zookeeper chart then, if people are already familiar with that one

[andimiller]

FYI, the zookeeper chart from bitnami (the one we depend on) also has this ability.

I've adjusted this to be closer how to the zookeeper chart works, with pdb as the config name, and exposing both minAvailable and maxUnavailable