Why is BIND logging "adb: grow_entries to <number>" and "adb: grow_names to <number>"?

  • Updated on Oct 30, 2018
  • Published on Nov 15, 2012
  • 1 minute(s) read
  • Suzanne Goldlust
  • Cathy Almond
 Prev Next

Question:

Since installing a new version of BIND, I'm getting the following informational messages on starting named. What do they mean?

18-Nov-2011 03:35:14.872 database: info: adb: grow_entries to 1531 starting
18-Nov-2011 03:35:14.874 database: info: adb: grow_entries finished

18-Nov-2011 03:36:21.908 database: info: adb: grow_names to 1531 starting
18-Nov-2011 03:36:21.911 database: info: adb: grow_names finished

Answer:

These are just diagnostic (informational) messages about the resizing of hash tables that are used by recursive servers. The feature (which improves run-time performance) was introduced in BIND 9.8.0. Depending on the loading on your server, you may also see them intermittently at other times - usually when there is a peak in query rates, or a spate of queries that are for names with a greater diversity than usual.

2993.    [func]        Dynamically grow adb hash tables. [RT #21186]

The ADB hash table sizes are never reduced, even if cache (including ADB) is emptied via rndc flush. There is no reason to decrease them, as the likelihood is that when repopulated following the flush, cache will once again reach a similar size.

Note on the new functionality
Currently named stalls other activity while this happens because the table being resized is accessed by multiple threads. This stall may or may not become externally noticeable, and if it does, we need to look at implementing a more complicated strategy to grow the hash table. However, we need the log messages now to correlate events and to be able to diagnose whether this is a problem.