You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Larger workloads (such as the usermem workload in benchmarking) do not currently run on GKE because the local snapshot directory (/run) is on a mount with < 1GB of space.
Max Smythe (@maxsmythe) This seems like a setup issue for GKE rather than for the project overall -- should we close this? Or should "/run" be configurable ?
This is a pretty big, hard-to-diagnose friction point at the moment that leads to mystery failures.
Also, documenting provider-specific quirks is a good way to smoke out oddities that could be streamlined by either making different choices (if there is a clearly better choice across cluster flavors) or providing simple configuration (if fix is a pain and everyone does something different).
Activity
thockin commentedon May 31, 2026
Max Smythe (@maxsmythe) This seems like a setup issue for GKE rather than for the project overall -- should we close this? Or should "/run" be configurable ?
azr commentedon May 31, 2026
Some docs on how to setup with GKE correctly or alternatively know GKE issues could help here.
maxsmythe commentedon Jun 5, 2026
+1 to GKE-specific docs.
This is a pretty big, hard-to-diagnose friction point at the moment that leads to mystery failures.
Also, documenting provider-specific quirks is a good way to smoke out oddities that could be streamlined by either making different choices (if there is a clearly better choice across cluster flavors) or providing simple configuration (if fix is a pain and everyone does something different).
Expose PostgreSQL migration settings in the Helm chart (agent-substra…