Long script is running out if memory

[ghost]

Hi Paul,

I have a realtively long measurement script ( multiple loadSlaves spawning and ending). I am measuring all RFC2544-defined packet sizes. The script always ends with error when generating traffic with 1500B packets(this is the last size in the list). I have observed 2 types of error:

1. The script exits with error code 134, no explanation.
2. The scripts exits with the following error: Could not run main script: not enough memory (I tried to add more hugepage, up to 8Gb but the error has remained)

I am using an i40e driven card. Do you have any idea what causes this problem? Maybe some kind of memory leak?

Thanks in advance, János

[emmericp]

Implicit mempool caching/recycling is unfortunately not enabled at the moment, i think the old rfc2544 scripts relied on that.

And I don't thinkt hat it's possible to re-implement that in general (bufs getting stuck and queues and thread-local caches), we've used some ixgbe-specific functions to explicitly drain queues. It was never reliable and is probably not a good idea that should be removed.

The fix for this, and other, applications is to explicitly pass a memory pool to the tasks.