#10180 NORM 1.5-sof: Journal hangs every five or six entry erasures
Zarro Boogs per Child
bugtracker at laptop.org
Mon Jun 21 02:58:53 EDT 2010
#10180: Journal hangs every five or six entry erasures
----------------------------------------+-----------------------------------
Reporter: Quozl | Owner: tomeu
Type: defect | Status: new
Priority: normal | Milestone: 1.5-software-later
Component: journal-activity | Version: Development build as of this date
Resolution: | Keywords:
Next_action: diagnose | Verified: 0
Deployment_affected: | Blockedby:
Blocking: |
----------------------------------------+-----------------------------------
Comment(by tomeu):
Replying to [comment:1 Quozl]:
> Diagnosis ... the stall corresponds to an unlink(2) call made by the
datastore service.
>
> {{{
> 02:54:15.303495
unlink("/home/olpc/.sugar/default/datastore/checksums/65033cfd0f41ac47b09e9f20c3bec62b/91491990-
5bd-42f0-8d79-093f63aa5534") = 0
> 02:54:21.379574
stat64("/home/olpc/.sugar/default/datastore/checksums/65033cfd0f41ac47b09e9f20c3bec62b",
{st_mod =S_IFDIR|0755, st_size=4096, ...}) = 0
> }}}
>
> So the stall is probably unavoidable in the current design of Sugar ...
Cannot be improved in the lower layers?
> this is a design problem in the Journal activity.
If by design you mean that these operations could be done async but aren't
currently, then yes. But making that change wouldn't mean too much work.
The current Journal and DS were tested on a XO-1 and the performance of
these operations was never so slow.
Just to be clear, every I/O operation should be done async if we want an
UI as responsive as possible, but it has a cost in development and you
need to expose the pending operations in the UI along with any errors,
eventual completion message, etc.
--
Ticket URL: <http://dev.laptop.org/ticket/10180#comment:3>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system
More information about the Bugs
mailing list