[ISSUE] (TEPHRA-146) Document that family max versions is required to be greater than one

1 view
Skip to first unread message

James Taylor (JIRA)

unread,
Oct 30, 2015, 2:33:22 PM10/30/15
to tephr...@googlegroups.com
James Taylor updated an issue
 
Tephra / Task TEPHRA-146
Document that family max versions is required to be greater than one
Change By: James Taylor
Summary: Confirm Document  that family max versions is  not  required to be greater than one
Add Comment Add Comment
 
This message was sent by Atlassian JIRA (v6.1.5#6160-sha1:a61a0fc)
Atlassian logo

James Taylor (JIRA)

unread,
Oct 30, 2015, 2:37:21 PM10/30/15
to tephr...@googlegroups.com
James Taylor commented on an issue
 
Re: Document that family max versions is required to be greater than one

From Gary Helmling:

Tables using Tephra for transactions must set max versions > the number of possible in-flight transactions (+ 1) at any given time. In practice, we normally do this by setting the table max versions to Integer.MAX_VALUE.

For user scans, Tephra will only return the most recent visible version, but for flush and compaction scanners it also needs to include all the all the other versions written by or visible to any in-progress transactions. The TransactionVisibilityFilter will take care of dropping any no longer visible versions of cells, so we rely on that to do the data cleanup instead of the HBase ScanQueryMatcher / ColumnTracker which does the built in version counting and cleaning.

So it's a requirement that the table max versions be > 1 and the safest setting is Integer.MAX_VALUE. Without that, it's possible that a flush or compaction would drop data (using the HBase built-in mechanisms) that is still visible to in-progress transactions, leading to incorrect results.

HBase only returns the minimum between family.maxVersions and scan.maxVersions for the number of cell versions in a user scan. This may or may not be an issue - just raising this JIRA at the request of [~poorna]. I discovered this when using the VisibilityLevel.SNAPSHOT_ALL capability, but am working around it by using a raw scan.

Poorna Chandra (JIRA)

unread,
Oct 30, 2015, 2:39:21 PM10/30/15
to tephr...@googlegroups.com

Poorna Chandra (JIRA)

unread,
Oct 30, 2015, 2:39:23 PM10/30/15
to tephr...@googlegroups.com

Poorna Chandra (JIRA)

unread,
Jan 12, 2016, 2:34:30 PM1/12/16
to tephr...@googlegroups.com
Poorna Chandra updated an issue
Change By: Poorna Chandra
Fix Version/s: 0.6.5
Fix Version/s: 0.6.4

Priyanka Nambiar (JIRA)

unread,
Feb 19, 2016, 4:42:27 PM2/19/16
to tephr...@googlegroups.com
Priyanka Nambiar updated an issue
Change By: Priyanka Nambiar
Fix Version/s: 0.7.1
Fix Version/s: 0.7.0
Reply all
Reply to author
Forward
0 new messages