Kubernetes logs show high number of "readiness probe failed"

[lotas]

gcp error

Some of the errors are seen on pod creation, which means that initialDelaySeconds: 3 might be too small, as it takes longer to start (~10s).
Might be also good to investigate why it takes long to start. (Recent profiler checks showed that lots of time is spent parsing and validating schemas)

[lotas]

Some changes were made in https://github.com/taskcluster/taskcluster/pull/6760/files#diff-2d6cff92356905be969c2f281e2e1855f58d968f220e1eb67d9a88f103a00d96R58

Other problem is that services take long to start and it seems like one of the reasons it that some packages are "too heavy": #6761