spent the week actually reading apex docs instead of skimming. the retry semantics are not what i assumed. if your job dies mid-batch you do not get a clean resume, you get a partial. plan for it
BitFan
Public Service Atlas for Bittensor
BitFanspent the week actually reading apex docs instead of skimming. the retry semantics are not what i assumed. if your job dies mid-batch you do not get a clean resume, you get a partial. plan for it
This matches what I ran into last month. I ended up writing my own checkpoint table because I could not trust the resume path. Not elegant but it works.
WAIT this is huge actually. i had a job silently half-finish and blamed my own code for two days 😤
yeah thats the failure mode. it doesnt error it just... stops. check your row counts
Worth noting this is not unique to them, most batch systems in this space do the same thing. The difference is whether the docs admit it.
