It's possible that local indexes could be allowed for atomic upserts, but global indexes are problematic (in that under load your cluster would probably die). The reason is that there'd be a cross RS call made for each row being atomically upserted. If the call hangs due to the RS hosting the data being down, it ties up the handler thread which can lead to a cascading effect when writes fail.
Not sure what your timeframe is, but support for Omid transactions is getting close to being committed. It's on the omid2 feature branch. They play well with HBase replication, so perhaps that's an option. On Fri, Jun 8, 2018 at 12:01 PM, Miles Spielberg <mi...@box.com> wrote: > This represents a serious shortcoming for our use case. We require some > level of same-row atomic update capability on tables we'd also like to have > global indices on. Our understanding is that running under Tephra doesn't > play well with HBase replication, which we've been intending to use for > disaster recovery. > > Our workload is read-dominated, so we think we're willing to accept the > row-locking and write time performance penalties. Is this something that > could be put under an option? Are there other significant shortcomings to > combining ON DUPLICATE KEY with global indices? > > PHOENIX-3925 mentions "service protection," but I'm clear on what that's > referring to. > > Miles Spielberg > Staff Software Engineer > > > O. 650.485.1102 > 900 Jefferson Ave > <https://maps.google.com/?q=900+Jefferson+Ave+%0D%0A%0D%0ARedwood+City+,+CA+94063&entry=gmail&source=g> > Redwood City, CA 94063 > <https://maps.google.com/?q=900+Jefferson+Ave+%0D%0A%0D%0ARedwood+City+,+CA+94063&entry=gmail&source=g> > > On Mon, Jun 4, 2018 at 7:14 PM, Sergey Soldatov <sergeysolda...@gmail.com> > wrote: > >> Yes, the documentation doesn't reflect the recent changes. Please see >> https://issues.apache.org/jira/browse/PHOENIX-3925 >> >> Thanks, >> Sergey >> >> On Fri, Jun 1, 2018 at 5:39 PM, Miles Spielberg <mi...@box.com> wrote: >> >>> From https://phoenix.apache.org/atomic_upsert.html: >>> >>> Although global indexes on columns being atomically updated are >>>> supported, it’s not recommended as a potentially a separate RPC across the >>>> wire would be made while the row is under lock to maintain the secondary >>>> index. >>> >>> >>> But the parser on 4.13.1 doesn't seem to agree: >>> >>> 0: jdbc:phoenix:thin:url=http://192.168.99.10> CREATE TABLE T1 (A >>> VARCHAR PRIMARY KEY, B VARCHAR); >>> >>> No rows affected (1.299 seconds) >>> >>> 0: jdbc:phoenix:thin:url=http://192.168.99.10> CREATE INDEX T1_B ON >>> T1(B); >>> >>> No rows affected (6.285 seconds) >>> >>> 0: jdbc:phoenix:thin:url=http://192.168.99.10> UPSERT INTO T1(A,B) >>> VALUES('hello', 'world') ON DUPLICATE KEY IGNORE; >>> >>> Error: Error -1 (00000) : Error while executing SQL "UPSERT INTO T1(A,B) >>> VALUES('hello', 'world') ON DUPLICATE KEY IGNORE": Remote driver error: >>> RuntimeException: java.sql.SQLException: ERROR 1224 (42Z24): The ON >>> DUPLICATE KEY clause may not be used when a table has a global index. >>> tableName=T1 -> SQLException: ERROR 1224 (42Z24): The ON DUPLICATE KEY >>> clause may not be used when a table has a global index. tableName=T1 >>> (state=00000,code=-1) >>> >>> 0: jdbc:phoenix:thin:url=http://192.168.99.10> >>> >>> Am I doing something wrong here? Is the documentation inaccurate? >>> >>> >>> Miles Spielberg >>> Staff Software Engineer >>> >>> >>> O. 650.485.1102 >>> 900 Jefferson Ave >>> <https://maps.google.com/?q=900+Jefferson+Ave+%0D%0A%0D%0ARedwood+City+,+CA+94063&entry=gmail&source=g> >>> Redwood City, CA 94063 >>> <https://maps.google.com/?q=900+Jefferson+Ave+%0D%0A%0D%0ARedwood+City+,+CA+94063&entry=gmail&source=g> >>> >> >> >
