[
https://issues.apache.org/jira/browse/HBASE-6768?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13543499#comment-13543499
]
Andrew Purtell commented on HBASE-6768:
---------------------------------------
bq. With many concurrent curl commands, the REST server will OOM very likely.
In this case, I don't think there are much we can do, right?
Yeah, REST has limits. To avoid the cost of HTTP overheads in cases of scanning
or multigets, REST has been designed to build the response -- issuing multiple
RPCs to the HBase cluster to do so -- and then send the response back to the
client all in one HTTP transaction. If a client request produces a really big
response, it has to fit in heap on the REST gateway. REST scanners do their own
batching to handle large datasets in chunks. For a given row request however we
can't divide it up, or request byte sub-ranges of values from the
RegionServers. First the Result (a row) is assembled from the Get or Scan
results inside the HBase client library. Then REST builds a model from the
Result, which ends up copying all of the data, because REST came before KV, so
its representation (Cell) predates it. Then the model is sent out by
Jersey/Jetty. The Result becomes a candidate for GC as soon as the model is
finished. The model is a candidate for GC as soon as Jersey finishes the
request.
> HBase Rest server crashes if client tries to retrieve data size > 5 MB
> ----------------------------------------------------------------------
>
> Key: HBASE-6768
> URL: https://issues.apache.org/jira/browse/HBASE-6768
> Project: HBase
> Issue Type: Bug
> Components: REST
> Affects Versions: 0.90.5
> Reporter: Mubarak Seyed
> Assignee: Jimmy Xiang
> Labels: noob
>
> I have a CF with one qualifier, data size is > 5 MB, when i try to read the
> raw binary data as octet-stream using curl, rest server got crashed and curl
> throws exception as
> {code}
> curl -v -H "Accept: application/octet-stream"
> http://abcdefgh-hbase003.test1.test.com:9090/table1/row_key1/cf:qualifer1 >
> /tmp/out
> * About to connect() to abcdefgh-hbase003.test1.test.com port 9090
> * Trying xx.xx.xx.xxx... connected
> * Connected to abcdefgh-hbase003.test1.test.com (xx.xxx.xx.xxx) port 9090
> > GET /table1/row_key1/cf:qualifer1 HTTP/1.1
> > User-Agent: curl/7.15.5 (x86_64-redhat-linux-gnu) libcurl/7.15.5
> > OpenSSL/0.9.8b zlib/1.2.3 libidn/0.6.5
> > Host: abcdefgh-hbase003.test1.test.com:9090
> > Accept: application/octet-stream
> >
> % Total % Received % Xferd Average Speed Time Time Time
> Current
> Dload Upload Total Spent Left Speed
> 0 0 0 0 0 0 0 0 --:--:-- 0:00:02 --:--:--
> 0< HTTP/1.1 200 OK
> < Content-Length: 5129836
> < X-Timestamp: 1347338813129
> < Content-Type: application/octet-stream
> 0 5009k 0 16272 0 0 7460 0 0:11:27 0:00:02 0:11:25
> 13872transfer closed with 1148524 bytes remaining to read
> 77 5009k 77 3888k 0 0 1765k 0 0:00:02 0:00:02 --:--:--
> 3253k* Closing connection #0
> curl: (18) transfer closed with 1148524 bytes remaining to read
> {code}
> Couldn't find the exception in rest server log or no core dump either. This
> issue is constantly reproducible. Even i tried with HBase Rest client
> (HRemoteTable) and i could recreate this issue if the data size is > 10 MB
> (even with MIME_PROTOBUF accept header)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
