[
https://issues.apache.org/jira/browse/HBASE-1744?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lars Francke updated HBASE-1744:
--------------------------------
Attachment: Hbase.thrift
This is my first stab at a completely new (again, based on Tim's experimental
idea) Thrift API. Quite a few points came up and I'd be interested in feedback:
* So far I have not included Filters in any way as there is no struct
subclassing in Thrift. So I'd need to come up with one data structure for all
Filters. I thought of using a list with struct entries that contain a
FilterType (enum) and a list or map of string options that would have to be
parsed. Probably not the best solution so if anyone has an idea go ahead
* A few things that are currently missing that I plan to add: Row lock, Scanner
caching, maxVersions, incrementColumnValue and checkAndPut
* A few things that I'm not yet sure about: The async version of createTable,
closeRegion (both from HBaseAdmin), getServerInfo from ClusterStatus, quite a
few methods from HTable that are not in HTableInterface (getRegionLocation,
getStartKeys, getEndKeys, getRegionsInfo, getRegionLocation
* What to do with autoFlush? I could easily add a method disableAutoflush or
something like that but what if flushCommit is never called? How does HBase
behave? I'd have to return a identifier for the table so that it can be reused
in subsequent calls (just like the scanner interface)
* I changed scanners to return longs before realizing that scanners timeout
after 60 seconds (default) so I'll probably change it back to ints
* Currently scanners are not cleaned up, I plan on adding a cleanup thread that
automatically closes/removes old scanners from the list
* Results are returned in TResult objects. You currently have to check if
isSetRow or if values.size == 0 - I am thinking of adding an boolen field
"empty" as this should be more reliable but as TResults will be sent quite
often every this would probably cause a noticeable overhead (I'm not sure how
well an optional field is handled in Thrift)
* Some structs have the "T" as a prefix to avoid name clashes with HBase (Get
-> TGet) but some don't (HTableDescriptor -> TableDescriptor, HColumnDescriptor
-> ColumnDescriptor), should I use the "T" everywhere to be consistent?
* The "delete" method had to be called something else in Thrift as "delete" is
a reserved keyword. Tim used "delet" but I think that's quite confusing (I
thought it was a type :) )so I opted for "deleteSingle" instead but I'm open
for suggestions
Well as I said: My first stab at this - I'm not very familiar with HBase so I
hope I got most of it right. If there are no major concerns I'll go ahead and
implement the rest of it (parts are done already) and upload a first patch.
Thanks for all the help on IRC (larsgeorge, dj_ryan, jdcryans and the others)!
> Thrift server to match the new java api.
> ----------------------------------------
>
> Key: HBASE-1744
> URL: https://issues.apache.org/jira/browse/HBASE-1744
> Project: Hadoop HBase
> Issue Type: Improvement
> Components: thrift
> Reporter: Tim Sell
> Assignee: Tim Sell
> Fix For: 0.21.0
>
> Attachments: Hbase.thrift, thriftexperiment.patch
>
>
> This mutateRows, etc.. is a little confusing compared to the new cleaner java
> client.
> Thinking of ways to make a thrift client that is just as elegant. something
> like:
> void put(1:Bytes table, 2:TPut put) throws (1:IOError io)
> with:
> struct TColumn {
> 1:Bytes family,
> 2:Bytes qualifier,
> 3:i64 timestamp
> }
> struct TPut {
> 1:Bytes row,
> 2:map<TColumn, Bytes> values
> }
> This creates more verbose rpc than if the columns in TPut were just
> map<Bytes, map<Bytes, Bytes>>, but that is harder to fit timestamps into and
> still be intuitive from say python.
> Presumably the goal of a thrift gateway is to be easy first.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
