Bind Auth responds slow during incoming XFR

Klaus Darilion klaus.mailinglists at pernau.at
Wed Mar 27 08:29:33 UTC 2019


Hello!

We have a problem with Bind [2] during incoming IXFR. When there is a
huge IXFR (ie 1,8GB tranferred in 15minutes [1]), the response time
heavily increases. Using dsc's newest "Reponse Time Indexer" we clearly
see that Bind answers slow:

Response Time   normal          during
Window          operation       IXFR
------------------------------------------------
1-10us          2
10-100us        16033
100-1000us      37768
1-10ms          180             21
10-100ms        43              179
100-1000ms                      748
1-10s                           2224
timeout         124             6930
------------------------------------------------
Sum             54150           10102

(The drop in total happens as the resolvers switch to other
authoritative nameservers)

The VM has 4 vCPUs which are all at 100% (99% user-space load)during the
incoming IXFR. The memory consumption doubled during the IXFR but there
is enough RAM in the server (no swapping)

So I have 2 theories:
a) While "patching" the in-memory zone the "IXFR-worker" somehow locks
the zone so that the "request-handler" has to wait and becomes slow.

b) The "IXFR-worker" eats all CPU so that there is no more CPU left for
the "request-handler".

So, what do Bind developers mean about my theories? Are there any ideas
to improve response time during IXFR? Ie. throttling CPU of the "IXFR
handler"? Were there related changed in 9.12 or 9.14?


Thanks
Klaus

[1] Transfer completed: 27733 messages, 12689911 records, 1789M bytes,
847.424 secs (2111887 bytes/sec)

[2] Seen on Ubuntu 14.04/16.04/18.04 aka Bind 9.9.5/9.10.3/9.11.3


More information about the bind-users mailing list