[jira] [Commented] (PHOENIX-3227) OFFSET is very slow at scale
[ https://issues.apache.org/jira/browse/PHOENIX-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15451521#comment-15451521 ] Ankit Singhal commented on PHOENIX-3227: Yes, the performance is expected. but I think for real use-cases , user will not pass such high offset(10M+) as query result sets generally ranges between 1-100k. For serial flat queries without filters, GuidePosts could be an option but the row count calculated are not actual and not available in realtime. So, IMO we can close this for now. > OFFSET is very slow at scale > > > Key: PHOENIX-3227 > URL: https://issues.apache.org/jira/browse/PHOENIX-3227 > Project: Phoenix > Issue Type: Sub-task >Reporter: Lars Hofhansl > > Perhaps this one is also expected since it is the client who has to count off > rows. In any case, we tried with an OFFSET of just 1000 and that took > 46s, regardless of how big the cluster at our disposal was. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (PHOENIX-3227) OFFSET is very slow at scale
[ https://issues.apache.org/jira/browse/PHOENIX-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15450425#comment-15450425 ] James Taylor commented on PHOENIX-3227: --- Not sure if [~an...@apache.org] has anything in mind to improve it, but if not we should close it. > OFFSET is very slow at scale > > > Key: PHOENIX-3227 > URL: https://issues.apache.org/jira/browse/PHOENIX-3227 > Project: Phoenix > Issue Type: Sub-task >Reporter: Lars Hofhansl > > Perhaps this one is also expected since it is the client who has to count off > rows. In any case, we tried with an OFFSET of just 1000 and that took > 46s, regardless of how big the cluster at our disposal was. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (PHOENIX-3227) OFFSET is very slow at scale
[ https://issues.apache.org/jira/browse/PHOENIX-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15450321#comment-15450321 ] Lars Hofhansl commented on PHOENIX-3227: Should I close this one [~giacomotaylor]? > OFFSET is very slow at scale > > > Key: PHOENIX-3227 > URL: https://issues.apache.org/jira/browse/PHOENIX-3227 > Project: Phoenix > Issue Type: Sub-task >Reporter: Lars Hofhansl > > Perhaps this one is also expected since it is the client who has to count off > rows. In any case, we tried with an OFFSET of just 1000 and that took > 46s, regardless of how big the cluster at our disposal was. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (PHOENIX-3227) OFFSET is very slow at scale
[ https://issues.apache.org/jira/browse/PHOENIX-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15450202#comment-15450202 ] James Taylor commented on PHOENIX-3227: --- [~an...@apache.org] > OFFSET is very slow at scale > > > Key: PHOENIX-3227 > URL: https://issues.apache.org/jira/browse/PHOENIX-3227 > Project: Phoenix > Issue Type: Sub-task >Reporter: Lars Hofhansl > > Perhaps this one is also expected since it is the client who has to count off > rows. In any case, we tried with an OFFSET of just 1000 and that took > 46s, regardless of how big the cluster at our disposal was. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (PHOENIX-3227) OFFSET is very slow at scale
[ https://issues.apache.org/jira/browse/PHOENIX-3227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15450197#comment-15450197 ] James Taylor commented on PHOENIX-3227: --- Yes, that's what I expected. FYI, we do push the OFFSET to the server when ever possible, but you still have to step through row by row (except on the server and then at least you don't return all the data back to the client). Also, the slowness is due to the query needing to be executed serially. > OFFSET is very slow at scale > > > Key: PHOENIX-3227 > URL: https://issues.apache.org/jira/browse/PHOENIX-3227 > Project: Phoenix > Issue Type: Sub-task >Reporter: Lars Hofhansl > > Perhaps this one is also expected since it is the client who has to count off > rows. In any case, we tried with an OFFSET of just 1000 and that took > 46s, regardless of how big the cluster at our disposal was. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
