It's a bad bug in that it's a silent corruption, so even if use of REST+JSON
isn't common it should be fixed soon. A quick patch that rejects empty values
is a start. Will add a "no I really mean it" flag. This would address 0.92 and
0.94. Then we switch to Jackson and a new JSON representation in 0.96. Sound
reasonable?
- Andy
On Jun 4, 2012, at 5:20 PM, Nick Dimiduk <[email protected]> wrote:
> Andy,
>
> Thanks for the follow-up. For what it's worth, I've been bitten by this
> "feature" by two separate projects this month. The only reason I thought to
> check field order was because it looked strikingly familiar.
>
> This is not a blocking issue for me, please prioritize accordingly. If no
> one else has noticed in 2 years, a documentation is more than sufficient.
>
> -n
>
> On Sun, Jun 3, 2012 at 2:54 PM, Andrew Purtell <[email protected]> wrote:
>
>> On Jun 3, 2012, at 11:51 AM, Nick Dimiduk <[email protected]> wrote:
>>> When I:
>>>
>>> $ curl -v -XPUT \
>>> -H "Content-Type: application/json" \
>>> http://localhost:9999/users/TheRealMT/info:password \
>>> -d '{
>>> "Row": [
>>> {
>>> "Cell": [
>>> {
>>> * "@column": "aW5mbzpwYXNzd29yZA==",*
>>> * "$": "NzBOQHJJIE4wIDcwdDBSMA=="*
>>> }
>>> ],
>>> "@key": "VGhlUmVhbE1U"
>>> }
>>> ]
>>> }'
>>>
>>> The logs show:
>>>
>>> 12/06/03 02:43:16 DEBUG rest.RowResource: PUT
>>> http://localhost:9999/users/TheRealMT/info:password
>>> 12/06/03 02:43:16 DEBUG rest.RowResource: PUT
>>>
>> {"totalColumns":1,"families":{"info":[{"timestamp":9223372036854775807,"qualifier":"password",
>>> *"vlen":16*}]},"row":"TheRealMT"}
>>>
>>> And the write is successful.
>>>
>>> But if I:
>>>
>>> $ curl -v -XPUT \
>>> -H "Content-Type: application/json" \
>>> http://localhost:9999/users/TheRealMT/info:password \
>>> -d '{
>>> "Row": [
>>> {
>>> "Cell": [
>>> {
>>> * "$": "NzBOQHJJIE4wIDcwdDBSMA==",*
>>> * "@column": "aW5mbzpwYXNzd29yZA=="*
>>> }
>>> ],
>>> "@key": "VGhlUmVhbE1U"
>>> }
>>> ]
>>> }'
>>>
>>> The log shows:
>>>
>>> 12/06/03 02:46:22 DEBUG rest.RowResource: PUT
>>>> http://localhost:9999/users/TheRealMT/info:password
>>> 12/06/03 02:46:22 DEBUG rest.RowResource: PUT
>>>
>> {"totalColumns":1,"families":{"info":[{"timestamp":9223372036854775807,"qualifier":"password",
>>> *"vlen":0*}]},"row":"TheRealMT"}
>>
>> See
>> https://issues.apache.org/jira/browse/HBASE-2475?focusedCommentId=13288259&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13288259
>>
>> and
>> https://issues.apache.org/jira/browse/HBASE-2475?focusedCommentId=13288265&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13288265
>>
>> Perhaps it's worth proceeding even though the JSON input and output of
>> HBase REST will most likely change? (And change to something cleaner
>> and saner.)
>>
>> We don't use the REST gateway ourselves but when we were looking at it
>> once, I recommended using the protobuf representation option.
>>
>> - Andy
>>
