Thank you Joan and Robert for your inputs.
We have tested with latest master of CouchDB and confirmed that the
problem still exists.
Regarding your questions:
- We have tested with single node. We tried both client and server on
the same machine and on different machines and it fails on both the cases.
- The only changes we made to the ini file are :
* to enable the logging level to debug.
* change bind_address to 0.0.0.0 to let CouchDB listen any
available IP address
* change port from 5984 to 5988 as 5984 is used by another
application in that machine.
[log]
level = debug
[httpd]
port = 5988
bind_address = 0.0.0.0
- We do not have any conflict version of the database in the system.
- We have collected the CouchDB logs and Wireshark traces from the
failing and passing cases (with delay while writing request body) and
uploaded to
https://drive.google.com/drive/folders/0BxTjd-f_AKG5RlpUVHl5RkRiUGs
Please review the logs and let us know whether they are good enough or
you need more logs.
Thank you,
Nasser Ebrahim
On 5/4/17 3:44 AM, Robert Samuel Newson wrote:
Agree with Joan, the most important thing is the log files.
If couchdb can send an error in the response, it will (a 413 or 404, etc, etc).
But if we've already started the response and _then_ encounter an error, we
can't send any useful information in the response, we have to close the
connection. When that happens, we log the error. You should find that the
request id you got matches something in the logs.
I expect it's a function_clause or case_clause, something of that nature, and
possibly indicating an unanticipated malformed request.
Logs pls.
B.
On 3 May 2017, at 20:42, Joan Touzet <woh...@apache.org> wrote:
Hi Nasser,
Thank you for the report.
Are you running against a single node or a clustered CouchDB 2.0
install? If clustered, how many nodes, and are they all running on
the same machine, or different machines? Have you changed any
settings in the ini files?
What sort of database do you have? Does it have any conflicted
versions in it?
Do you have any CouchDB logfiles from when the error occurs? Do any
of them show anything useful? You can set the logging level to debug
to gather additional information.
Please don't email logfiles directly to this list; you can share them
with a service like gist.github.com, pastebin.com or paste.apache.org
instead.
Finally, have you tried running against our current master rather
than the released 2.0 version? We've fixed a lot of bugs since then,
and it's possible this bug has already been resolved as the result of
an unrelated change.
-Joan
----- Original Message -----
From: "Nasser Ebrahim" <enas...@linux.vnet.ibm.com>
To: dev@couchdb.apache.org
Sent: Wednesday, 3 May, 2017 1:48:06 PM
Subject: Truncated response when POST a _changes query
Hello,
While doing the Cloudant swift test, we are getting truncated response
when POST a _changes query to the CouchDB with document ID [
<http://docs.couchdb.org/en/2.0.0/api/database/changes.html>http://docs.couchdb.org/en/2.0.0/api/database/changes.html].
We are getting the failure very frequent while doing the test from a
swift client on Linux with couchDB 2.0 as server. We compared the TCP
stream of the passing and failing case and the request is exactly the
same. Hence, we believe there is something going wrong while processing
the request on the CouchDB side as we are getting the truncated response.
Another interesting observation is that if we introduce a small delay
(sleep) before writing the request body on the swift client side, the
test is passing (the response from CouchDB is proper). Hence, we think
this could be a timing related issue on the CouchDB side.
While doing the same Cloudant swift test from Mac OS, we are observing
the failure very rarely. We believe it could be the change in timing
which hide the issue similar to when we introduce the delay while
testing on Linux.
The response from the CouchDB has three chunks. The first chunk is a
standard text {"results”:[, the second chunk is the actual response and
the last chunk is the standard stream terminator sequence. In the
failing case, we are getting only the first chunk. Hence, it seems the
failure occurred while processing the response on the CouchDB side.
We have taken the CouchDB trace and Wireshark trace from the server side
and we could confirm that the request is exactly the same between the
passing and failing case where as the response is truncated on the
CouchDB side during the failure.
Please let us know whether you are aware of any such issues on the
CouchDB side and what diagnostic documents are required for you to do
the analysis.
Thank you,
Nasser Ebrahim
