Task #25 — Wrong serial number matching for slave domains
Attached to Project — PDNS OpenDBX| Bug Report | Low | ||
| Backend / Core | Normal | ||
| Unconfirmed | 1.0 | ||
| No-one | Undecided | ||
| All |  |
||
|
We have two powerdns 2.9.22 with opendbx/sqlite3 here. Each of the server has two master domains and should have two slave domains from the other server, which is set as supermaster. We use the auto serial feature. One server is running gentoo, the other one is running debian. Both on amd64 platform. The master domains and the first slave domain works well. Here comes the issue: with the second slave domain, the serial id comparisons start to fail (see below) and one of the slave domain is dropped. See example log line (1st.domain has serial 51 on master, 2nd.domain has serial 49): Domain 2nd.domain more recent than master, our serial 102 > their serial 49 Note, that 2nd.domain's serial number is compared to 1st.domain's serial number, shift one bit to the left! (It's always this way). Can't there be any pointer/cast bug? |
|||
Comments (4) | Attachments (2) | Related Tasks (0/0) | Notifications (1) | Reminders (0) | History |
Comment by - Tuesday, 1 Dec 2009, 8:03am
Could you provide my your configuration, database structure (including the triggers) and some log output which shows the problem? Did you compiled the code yourself or which distribution package do you use?
Comment by Vladimir Linek - Wednesday, 2 Dec 2009, 1:29pm
Dump and conf attached.
It's from the gentoo server -- distribution ebuild used.
As you can see from the database dump, zagamma.cz and py.cz have master records here, stral.cz and satral.cz (testing domain) are slaves and kicking each other from the dns server.
Comment by Norbert Sendetzky - Thursday, 3 Dec 2009, 10:20pm
Could you also attach the log file where I can see what's happening?
Bert fixed an issue regarding lost slave updated a week ago which might be also the one you are stumbled upon. Could you try a version with the patch (http://wiki.powerdns.com/trac/changeset/1444) applied to?
Comment by Vladimir Linek - Tuesday, 8 Dec 2009, 3:19pm
I've attached log file from the "other" server (prg-01-home, ubuntu, patched).
First, we removed all slave records. Then applied the patch. And after a server restart we sent out the notifies.
As you can see, the issue is still there, and all records for one domain (zagamma.cz this time) were dropped.