Le 6/01/2017 à 12:49, Santiago Vila a écrit :
> But I still don't understand why it should fail. I measure the
> allocated memory by looking at Committed_AS in /proc/meminfo, and this
> package has never required more than 900 MB (less than 768MB + 1GB).
> Maybe it's the stack (not the heap, measured by Committed_AS) what
> my autobuilders had not enough?
I don't know, a wild guess would be that the JVM didn't provision enough
heap due to OS constraints (the initial heap size depends on the memory
available), and the usage of the heap grew too quickly and hit the max
before the JVM had the time to borrow more memory from the OS.
> How could I ensure that this does not happen again in the general case?
I think allocating more memory to the builder should address this issue.
> There is also another problem: I have at least one *successful* build
> containing an error like that (see attach).
Maybe sometimes the JVM does manage to resize the heap before the overflow.
> Is this not a violation of Policy 4.6? (That's the rule saying that
> whenever there is an error, the build should stop).
Maybe the build would have stopped if not killed after 60 minutes though.
This is the maintainer address of Debian's Java team
debian-j...@lists.debian.org for discussions and questions.