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

Re: Replication oddity in 2.3.19

> Update master, nothing happens (on link traffic).  Interrogate slave: no
> change, even after several minutes.  Interrogate master, then interrogate
> slave: see the change appear.  Perfectly reproducible.
> Some weird caching problem?  Anyone else seen this?

I'm pretty sure that I've seen syncrepl behave in ways that I don't like.
The last weird case I confirmed was an ADD. It went through on the master,
hit 2/3 slaves within minutes, and then hit the last slave...22 hours
later. Seeing as they share the same config file over NFS, this
seems...questionable. I was looking into this heavily and was hoping to
throw in more debugging code into the session log functions (do you have a
session log configured? I'm kind of suspecting it, but no hard
evidence...hence the need for debugging!) but then other projects took
priority...not sure when I'll get back to it.

The real win would be "perfectly reproducible" in terms of, say, test017
(edit the config if necessary). I was nowhere close to that... But still,
if you're able to instigate this easily, a couple -d -1 dumps might help.
Mine showed send_ldap_intermediate never getting called...of course the
question is why!