[MB-8461] [Windows] Stat errors?: Swap usage showing 400% Created: 12/Jun/13  Updated: 22/Jul/13  Resolved: 18/Jul/13

Status: Closed
Project: Couchbase Server
Component/s: UI
Affects Version/s: 2.1.0
Fix Version/s: 2.2.0
Security Level: Public

Type: Bug Priority: Major
Reporter: Abhinav Dangeti Assignee: Sriram Melkote
Resolution: Won't Fix Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified
Environment: Windows 2.1.0-717-rel

Attachments: PNG File Screen Shot 2013-06-12 at 5.16.35 PM.png     PNG File Screen Shot 2013-06-12 at 5.45.44 PM.png     PNG File Screen Shot 2013-06-12 at 6.07.31 PM.png    

Page File size allocated: 4176MB
Swap usage as per UI on that server (in the buckets tab) is at around 200MB

But the Swap usage in percentage on the servers tab keeps increasing (snap shot at 400% attached).
(Noted this number during rebalance, never came back down after rebalance though)

The task manager process performance screenshots show that memory usage is at around 80-90%, and there are a few hard faults observed.

Attaching memcached and erlang' process dumps (from task manager, if this helps):

Comment by Sriram Melkote [ 13/Jun/13 ]
On Gerrit:

The fix for 2.1.0 to this is to calculate the current page file extent using commit limit minus physical memory. This multiplied by usage fraction gives accurate estimate of actual page file size (and this is different than what appears under "Page File" of task manager, which is maximum potential usage).

However, for next minor release, we must replace the Swap Percentage Use indicator with either the ratio of Page File Size to Physical RAM, or just total page file size.

This is because page file size can be dynamic (and is so by default on Windows) and vary widely in size on identical machines. On Windows, where past maximum memory requirement largely determines the maximum page size. So 10% page file usage on a machine with a big page file may represent a bad memory situation, while it may be perfectly normal elsewhere.

Comment by Ravi Mayuram [ 14/Jun/13 ]
Deferring to 2.1.1.
Comment by Sriram Melkote [ 18/Jul/13 ]
Alk has rolled everything back to portsigar, so this bug (though fixed) is no longer applicable.
Generated at Sun Nov 23 13:07:56 CST 2014 using JIRA 5.2.4#845-sha1:c9f4cc41abe72fb236945343a1f485c2c844dac9.