Just wanted to document a scenario to look out for...
# generalb
bitter-ability-32190
06/19/2023, 10:15 PMJust wanted to document a scenario to look out for and maybe mark something for Pants improvement.
Someone was complaining their PR took 17 minutes to pass CI without touching code. We use a remote cache, and every CI runner uses the same AMI, so this is a huge red flag.
After debugging I found out that I could reproduce if I wiped the PEX named cache. Turns out an sdist had an irreproducible build. Every time you build it, you get a different fingerprint.
This is a gotcha for sure. But also highlights how we likely should rely on PEXs cache less, and rely on our own more (that's hand-waivey as hell, but also true)
👍 1
bitter-ability-32190
06/19/2023, 10:16 PMWe'll be switching to an internal PyPI mirror and uploading wheels in the meantime.
Probably a good idea regardless, but still
p
proud-dentist-22844
06/20/2023, 3:14 AMIirc John filed an issue about the cache key being not quite correct in pants...
proud-dentist-22844
06/20/2023, 8:03 AMAh. I was thinking of an orthogonal issue. https://github.com/pantsbuild/pants/issues/18250
So, not quite the same, but it should probably be cleaned up to fix the hash for `Collection`s and
CoarsenedTargetsproud-dentist-22844
06/20/2023, 8:05 AMPEX tries very hard to make things reproducible (same fingerprint). What is causing the fingering to change?
b
bitter-ability-32190
06/20/2023, 10:51 AMThe sdist produces a different binary blob. Nothing PEX can do to stop that. (Think embedding a timestamp into the wheel)
🤦 1
bitter-ability-32190
06/20/2023, 11:02 AMI think https://github.com/pantsbuild/pex/issues/1423 is likely related
2 Views