Hi Mauricio,

this sounds indeed rather bad. I would very much like to get to the bottom 
of this. First I need some more information about your setup, in particular:

  (1) Are you running a single server or cluster setup? (Port 8529 suggests 
a single server, but I cannot really know)
  (2) Are you running on bare metal, and if so, what OS are you using? 
(please include the version)
  (3) Are you running in the cloud, and if so, what provider with which 
instance type? (for AWS, in particular, we also need to know the attached 
storage, for example, EBS can have rather severe IOPS throtteling, which 
can cause harm)
  (4) How much memory and what type of disk have you attached?
  (5) Which version of ArangoDB are you using?
  (6) How large are your import files, and how large (approximately) are 
the individual records?
  (7) Can you please send the exact `arangoimp` commands you use? Are you 
running them on the same machine or over the network?
  (8) Do you see any abnormal CPU load during the import?
  (9) Are you able to run `iotop` to see the I/O throughput during the 
import (you might need to use `sudo` to get root privileges)
  (10) Are you running the UI frontend during the import? (try to not run 
it, just as a wild guess!)
  (11) Is there anything abnormal in the database logs?
  (12) Is the machine use by anything else at the time of the import?

If you could at least provide answers to a part of these questions, then I 
could more easily try to replicate (and then solve) your problem. You can 
also send me information privately by email via
  max at arangodb.com

As to your question: Currently, the lock timeout cannot be configured, it 
is fixed at 15s, which might be too small. However, I do not see anything 
in the load you produce which would involve an operation that has to wait 
for a lock. Therefore, I suspect that there is a more fundamental problem 
than the timeout time. What you could change is the batch size in 
`arangoimp` (try to make it smaller than the default 16777216 bytes).

Note that the truncate operation is transactional and is therefore much 
slower than dropping the collection and recreating it.

As mentioned earlier, I would very much like to get to the bottom of these 
nasty problems.

Best regards,
  Max.

-- 
You received this message because you are subscribed to the Google Groups 
"ArangoDB" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to