[
https://issues.apache.org/jira/browse/TRAFODION-860?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14721174#comment-14721174
]
Suresh Subbiah commented on TRAFODION-860:
------------------------------------------
Anuradha (anuradha-hegde) on 2015-01-07
Changed in trafodion:
assignee: nobody → Matt Brown (mattbrown-2)
milestone: none → r1.0
Trafodion-Gerrit (neo-devtools) wrote on 2015-01-09: Fix proposed to dcs
(master) #1
Fix proposed to branch: master
Review: https://review.trafodion.org/929
Changed in trafodion:
status: New → In Progress
Matt Brown (mattbrown-2) on 2015-01-09
Changed in trafodion:
status: In Progress → Fix Committed
Trafodion-Gerrit (neo-devtools) wrote on 2015-01-10: Fix merged to dcs (master)
#2
Reviewed: https://review.trafodion.org/929
Committed:
https://github.com/trafodion/dcs/commit/a5c766c7aef13584706810636a77f4eda712baa4
Submitter: Trafodion Jenkins
Branch: master
commit a5c766c7aef13584706810636a77f4eda712baa4
Author: matbrown <email address hidden>
Date: Fri Jan 9 19:45:13 2015 +0000
Disable mxosrvr port map
Closes-Bug: #1408454
In order to improve mxosrvr start time performance 2 new properties were
added:
dcs.server.user.program.port.map.timeout.seconds = 60
dcs.server.user.program.port.bind.timeout.seconds = 30
They are passed to mxosrvr at start time. The DcsMaster would create a
port map in the servers/registered znode so mxosrvrs could identify
their DcsServer instance/child id and discover the TCP/IP port to use.
During performance testing, it was found that because DCS/MXOSRVRs usage
of ports in the ephemeral range DCS couldn't guarantee dedicated use of
a given range of ports. Setting the 2 properties to -1 disables port
mapping and returns to original herding style port identification
dcs.server.user.program.port.map.timeout.seconds = -1
dcs.server.user.program.port.bind.timeout.seconds = -1
Change-Id: I0497bfbc58878378d67a8cba9293708e88dc271e
Guy Groulx (guy-groulx) wrote on 2015-01-14: #3
Fixed as seen in git150114.
Changed in trafodion:
status: Fix Committed → Fix Released
> LP Bug: 1408454 - DCS: Can not start more than 200 mxosrvrs across cluster.
> -----------------------------------------------------------------------------
>
> Key: TRAFODION-860
> URL: https://issues.apache.org/jira/browse/TRAFODION-860
> Project: Apache Trafodion
> Issue Type: Bug
> Components: connectivity-dcs
> Reporter: Guy Groulx
> Assignee: Matt Brown
> Priority: Blocker
> Fix For: 1.0 (pre-incubation)
>
>
> Installed git150107 onto our system.
> We have dcs servers file configured to start 128 servers on each of our nodes
> (10 nodes in total).
> With the dcs-150107 version, we got many cores.
> echo /local/cores/1005/core.1420648319.n013.25176.mxosrvr
> echo
> [New Thread 25176]
> [New Thread 25834]
> [New Thread 25937]
> [New Thread 25936]
> [New Thread 25833]
> [New Thread 25832]
> Core was generated by
> `0:63:21063:10:64:21064:10:65:21065:10:66:21066:10:67:21067:10:68:21068:10:69:21'.
> Program terminated with signal 11, Segmentation fault.
> #0 0x00007ffff4d48d5f in ?? ()
> #0 0x00007ffff4d48d5f in ?? ()
> #1 0x00000000004dd2ed in ?? ()
> In testing various scenarios we found that we could start on 2 nodes with 100
> servers each. But at 128, the ports would not connect correctly.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
