I don't understand this suggested change. On the one hand, it sounds like you are referencing a guc that controls the maximum number of vsegs per segment which I don't believe there is one.
The summary suggests that you want to overload hawq_rm_nvseg_perquery_perseg_limit to not only be for a session but also set the maximum for the cluster. I don't agree with this at all. There is another guc hawq_rm_nvseg_perquery_limit which sets the maximum number of vsegs for the entire cluster. This is rather confusing especially for a cluster where compute and storage are separate and the use case is for a growing and shrinking cluster. I would rather have a guc for the maximum number of vsegs per segment. This is something you referenced but again, I don't think such a guc exists. Jon Roberts On Thu, Dec 22, 2016 at 1:27 AM, Xiang Sheng (JIRA) <[email protected]> wrote: > Xiang Sheng created HAWQ-1231: > --------------------------------- > > Summary: RM should error out when nvseg in statement level is > greater than the guc for nvseg per seg. > Key: HAWQ-1231 > URL: https://issues.apache.org/jira/browse/HAWQ-1231 > Project: Apache HAWQ > Issue Type: Bug > Components: Resource Manager > Reporter: Xiang Sheng > Assignee: Ed Espino > Fix For: backlog > > > If user set a larget nvseg in statement level, the nvseg is greater than > the guc nvseg per seg, rm won't report error. It will allocate according to > nvseg per seg. But this cannot satisfy statement level nvseg. RM will > timeout at last. This behaviors is not reasonable. We should error out > directly and tell user if he wants the nvseg in statement, he must change > the guc ( hawq_rm_nvseg_perquery_perseg_limit ) accordingly. > > > > -- > This message was sent by Atlassian JIRA > (v6.3.4#6332) >
