https://bz.apache.org/SpamAssassin/show_bug.cgi?id=8033
--- Comment #3 from Sidney Markowitz <sid...@sidney.com> ---
The frustrating stage I have gotten to now, and a speculation on a fix:
I'm still refining adding debugging logging to narrow down where it happens,
with 24-48 hour turnaround on results, but I have shown that the temporary
directory created in /dev/shm disappears all of a sudden with the exact place
in the code that it does varying a bit between runs, happening in either the
second or third call to sa-learn
My speculation is that this particular set of test machines have something that
reaps files in /dev/shm too aggressively.
The temporary directory is created using the lines
# Try /dev/shm as it's likely memdisk, otherwise SQLite is sloow..
my $dbdir = tempdir("bayessql.XXXXXX", DIR => -w "/dev/shm" ? "/dev/shm" :
"log");
The comment is correct. On my Ubuntu virtual machine running in VirtualBox on
my MacBook Pro, the test takes about 23 seconds to run using /dev/shm and
2min:30sec if that line is changed to always use "log".
The question I have is whether that difference in time for that test is worth
it if some machines that the test runs on have a /dev/shm that is writable but
is not set up suitably for use for temporary files.
If it turns out that never using /dev/shm for the files allows the test to run
on all the CPAN test machines instead of failing on a few, at the cost of the
test taking two minutes longer to run on machines that have a /dev/shm, is that
the correct fix? (It will take a couple days to get the feedback to see if that
change does cause the failures to go away).
Note that this only has to do with the one test. It has nothing to do with
anything that runs in actual SpamAssassin. By our guidelines I wouldn't even
have to get a vote for RTC. But I am asking for thoughts and comments on this
issue.
--
You are receiving this mail because:
You are the assignee for the bug.
