<@UB2J9BQA0>: re: <https://github.com/pantsbuild/p...
# development
w
the bug (afaict) is that cache entries are written for failed processes
h
which we can't control, that's up to each RE server implementation
w
sure: but that’s the impact of that flag: the server won’t check the cache
h
w
the first PR is correct, yea
h
sure: but that’s the impact of that flag: the server won’t check the cache
yeah, if
--no-remote-cache-read
, then both the server will be instructed to not do its own internal remoet cache lookup, and also our own RemoteCache commandrunner won't do a lookup. Otherwise, it's possible RE server will still look it up because we cannot control the behavior of that server
w
Otherwise, it’s possible RE server will still look it up because we cannot control the behavior of that server
we can: that’s what the flag is for, heh
h
which is why I'm now setting the flag? which was not set before. I'm confused
w
basically, i think that the flag should always be set in
remote::CommandRunner
, regardless of `cache-read`/`cache-write`, because we are doing our own cache reads/writes
h
oh, yeah that could make sense
w
i think that it needs to be to actually fix the bug
h
why would it need to be? seems only like a possible performance optimization to avoid duplicated work of both the server and Pants doing the same cache lookup
w
because the server doesn’t have our failure logic
h
like ProcessCacheScope?
w
right.
h
yeah, that makes sense
okay
w
thanks: confusing.