[LNT] r249026 - Add support for storing hash of test binaries.

Tobias Grosser via llvm-commits llvm-commits at lists.llvm.org
Mon Oct 12 10:03:15 PDT 2015


On 10/12/2015 06:25 PM, Chris Matthews wrote:
> One thing I do is have the builds store their report.json file, so when the submissions get mucked up I can manually fix and then do a "lnt submit". This has saved hundreds of results from being lost on Green Dragon. Better still, when I bring up a test instance, I can blast the last few months of results into it, so I have real data.

I remember the llvm.org/perf server also stored the report.json files.

> I think there is a constant danger of problems like this when our bots all work on ToT LNT.  I don't know if doing something more sane is worth the effort though? Maybe upgrade llvm.org perf at the start of every month? At least you'd know when it will fail.

I personally would be fine with auto-upgrading, assuming we agree we 
want to keep the server working and quickly revert/fix commits if
issues are encountered. This works for LLVM and seems to be a reasonable
model for LNT as well.

Btw. did you manage to update LNT to unbreak the current builders?

Best,
Tobias


More information about the llvm-commits mailing list