With a recent upgrade to Unbound 1.7.3, we have been noticing instances of the following errors in our logs:
Could not generate request: out of memory
mem error generating DS request
System memory stats are fine, open file descriptors are within limits, used sockets are within limits. We cannot correlate these errors with any problematic symptoms.
We want to be thorough and we want to be purposeful. Should this error be ignored, or is it prudent to give unbound more memory for DS requests? Which configuration options apply?
With a recent upgrade to Unbound 1.7.3, we have been noticing instances
of the following errors in our logs:
What must be the cause is not a memory failure, it prints the wrong
thing, this code path is also excercised for memory failures, but in
this case I think the cycle detection made it fail. This is the cycle
detection in that DNS lookups need other DNS lookups to succeed and a
cycle is detected in this. Unbound stops it. This cycle detection is
printed on verbosity level 4. And then this error is generated.
Not sure how this is caused, eg. a qtype DS lookup is needed for ... for
itself.
It can be harmless to ignore, but I should improve the error reporting.
Unbound can try other options after detecting the cycle, and if those
succeed, the problem is solved.