No space left on device error about gg HOT 4 CLOSED

I see that lambda function has garbage collection in the beginning of the handler but sometime it is not enough.

You're totally right, but now I see that it is not enough. We put all the objects in /tmp/_gg directory and one solution would be getting rid of that altogether (we only write to this directory and /tmp/thunk-execute.*), but it's not the best decision, because it's possible that some of those blobs can be reused for the current thunk, and we would have to re-download them again.

I just pushed a change (5331864) that cleans out /tmp/_gg, only if the Lambda is out of space. It should fix this.

AWS Lambda Execution Context is something that lives inside AWS and does not have any external control capabilities.

This is my understanding, too.

It may help to transfer the tmp cleaning code to the end of the function to avoid situation when the N-1 call "clogged up" memory.

gg execute cleans up the execution directories itself after the task is done; main.py:35 is there to make sure we don't have any leftovers (e.g., in the case that the Lambda dies mid-job).

from gg.

The cause of the problem is described here.

I see that lambda function has garbage collection in the beginning of the handler but sometime it is not enough. I saw this error once when all executions of my function failed (there was an error on my side). A simple solution is to just wait for AWS to drop the execution context (~ 10 minutes).

from gg.

Yeah, I noticed this went away after a while. I don't suppose there is a way we could get gg to flush/remove the context if it picks up this error?

from gg.

AWS Lambda Execution Context is something that lives inside AWS and does not have any external control capabilities. Moreover, a full description of the Execution Contex lifecycle is also unclear. We only know that it exists and is something like a cache, to support the "warmed up" lambda functions.
It may help to transfer the tmp cleaning code to the end of the function to avoid situation when the N-1 call "clogged up" memory. And only @sadjad can say whether it makes sense.

from gg.