Open droazen opened 7 years ago
For @tedsharpe
@droazen I have a PR for gatk-bwa-mem that adds a footer to the image file so that we can test integrity. I also added code to test every (I think) call that returns an error indication, and pass this info up the chain. Could you review the PR or delegate, please?
@droazen Ditto for gatk-fermi-lite: there's a new PR that adds some error handling. However, that artifact is hooked up to Travis, which fails. So I'll probably need some help from @lbergelson to get that sorted out -- gradle is missing an assemble verb. Can you review the PR or assign a reviewer, please?
@tedsharpe I'm going to nominate @TedBrookings to do a first-pass review to check the system calls, as based on my conversation with him the other day I think he might be a good person to review. I'll do a review pass as well to have a look at the CRC stuff, etc.
We are currently plagued with cryptic intermittent failures coming from the BWA and FML bindings in Travis CI. These usually manifest as a simple "exited with code 137" (ie., killed by signal 9) error, but sometimes we get an explicit segfault or out-of-memory error. Examples:
The underlying issue in these cases is likely either "out of memory" or, perhaps in the case of the seg faults, "file not found" or "malformed file", but we could greatly improve our ability to interpret Travis failures if we were more careful about checking return values from system calls. Eg., in the function below from the BWA bindings we could check the return values of the
mmap()
andcalloc()
calls, and die with an appropriate error message if they fail: