[Date Prev][Date Next] [Chronological] [Thread] [Top]

(ITS#4423) syncrepl'd MODs cause DELs



Full_Name: Aaron Richton
Version: 2.3.20
OS: Solaris 9
URL: https://www.nbcs.rutgers.edu/~richton/nonpresent/
Submission from: (NULL) (128.6.31.135)


(unlike the previous ITS, this one almost has a bit of proof)

I've been attempting to come up with good reproduction for the issues I've been
seeing with syncrepl'd MODs causing be_deletes on the slaves. So far, I've
failed to come up with a consistent reproduction. However, my best attempt has
succeeded in creating runs where syncrepl_del_nonpresent hits despite a test
where only ADD and MOD are sent to the master! As I started composing this ITS,
I set the test in a continuous loop. So far, del_nonpresent is in slapd.2.log on
3/15 runs. I'm not sure if this is some very rare timing issue, a cosmic ray, or
what, but it certainly looks very similar to what I'm seeing in production.

I've uploaded the tests/data files I've modified, a script to drive my test
(based on the glue-syncrepl test), and a tarball showing a good run and an off
run that hdb_delete's Dorothy Stevens. Surely slapd should be modifying instead
of deleting here...I just can't figure out what's triggering it with something
so rare. Hopefully the -d -1 makes more sense to someone out there...