Opened 16 years ago

Closed 16 years ago

Last modified 15 years ago

#656 closed Defect (fixed)

Project server crashed - client messed the computer location.

Reported by: Pepo Owned by: davea
Priority: Major Milestone: 6.2
Component: Client - Scheduler Policy Version: 6.2.1
Keywords: Cc: Pepo

Description

SETI's BOINC database server crashed today. Few minutes later my host contacted it:

19:48:15 [SETI@home] [sched_op_debug] Starting scheduler request
19:48:15 [SETI@home] Sending scheduler request: To fetch work.  Requesting 8 seconds of work, reporting 0 completed tasks
19:48:20 [SETI@home] Scheduler request succeeded: got 0 new tasks`
19:48:20 [SETI@home] [sched_ops_debug] Server version 601
19:48:20 [SETI@home] Message from server: Server can't open database
19:48:20 [SETI@home] Project requested delay of 3600.000000 seconds
19:48:20 [SETI@home] New computer location: 
19:48:20 [---] General prefs: from SETI@home (last modified 13-May-2008 10:27:28)
19:48:20 [---] Host location: none
19:48:20 [---] General prefs: using your defaults
19:48:20 [---] Reading preferences override file
19:48:20 [---] Preferences limit memory usage when active to 1023.02MB
19:48:20 [---] Preferences limit memory usage when idle to 1841.44MB
19:48:20 [---] Preferences limit disk usage to 0.00GB

sched_reply_SETI.xml contained just following:

<scheduler_reply>
<scheduler_version>601</scheduler_version>
<master_url>http://setiathome.berkeley.edu/</master_url>
<request_delay>3600.000000</request_delay>
<message priority="low">Server can't open database </message>
<project_name>SETI@home</project_name>
</scheduler_reply> 

The host was "thrown off" from its venue and related settings and the mess started: the machine was constantly suspending/resuming CPU throttling, switched off network and effectively nothing was being crunched anymore.

The client should be resistent against such empty returned data.

Change History (2)

comment:1 Changed 16 years ago by davea

Resolution: fixed
Status: newclosed

(In [15230]) - client: If a scheduler reply doesn't include <host_venue>

(as is the case if its DB is down, or it has no work and skips the DB check) don't overwrite the host_venue on the client. Fixes #656, I hope (Rom: please port)

comment:2 Changed 16 years ago by Nicolas

So the fix in #578 was useless, because the client still deleted the venue even if the server returned <project_is_down/>?

Also, there is an identical problem with team name, see #417 and #633.

Note: See TracTickets for help on using tickets.