Web agent comes with 16gb of ram and no swap (last time I checked) And when I try to load some financial data with pandas it just broke and agent replied with ‘Shell fails with Pod … failed to start: terminated’ and the VM totally broke and the session cannot continue. I did some digging and pinned down that it’s 99% a OOM error. And I tried to ask the agent to add some swap, but it does not support. I managed to get around it by tweaking the code. If you can fix this either by allowing swap or at least tell that it’s OOM when it broke that would be very helpful.
Steps to Reproduce
Use the python pandas package and ask it to load more data than 16gb can handle
Expected Behavior
‘Shell fails with Pod … failed to start: terminated’ and the VM totally broke and the session cannot continue
Hey, thanks for the detailed report and for digging into the OOM yourself, that really helps.
On the two parts:
The unclear error. You’re right. When the VM hits the memory limit, instead of a clear out-of-memory message you see an opaque Pod ... failed to start: terminated, and the session won’t continue. That shouldn’t happen, and it’s not something in your setup. I’ve passed this to the team and we’re tracking it. I can’t share a timeline yet, but I’ll reply here if there’s an update.
RAM and swap. Right now the web Cloud Agent defaults to 16GB RAM with no swap. That’s the current behavior, and the agent can’t enable swap on its own since it needs a pod-level config. Bigger VMs aren’t self-serve yet. I also passed along your request for a configurable VM size as feedback.
Until this is fixed, for heavy pandas loads it helps to:
Read data in chunks using chunksize= in read_csv and aggregate per chunk.
Downcast dtypes like float64 to float32, use categorical columns, and use parquet instead of CSV where possible.
Move the heaviest step to your own machine or to a local CLI agent if you know it’ll exceed 16GB.
Looks like you already worked around it by changing the code. If you hit it again, message me and we can take a look together.