[Pacemaker] attrd does not update crm_master

[hj lee]

On Mon, Jan 11, 2010 at 1:22 PM, hj lee <kerdosa at gmail.com> wrote:

> Hi,
>
> I have a situation crm_master does not update a master score. I increased
> log level at attrd. And I found that current value is 20000 but stored value
> is 5. I don't know how attrd ends up this state. But attrd ignores this
> update, so a master score does not change at all. I think the code should
> check the stored value, not current value in memory. As long as the stored
> value is different, then it should update. Should I create a bug?
>
> Thanks
> hj
>
>
> 2010-01-11 09:37:16.425570 copper2-attrd: [4649]: debug: debug3:
> log_data_element: attrd_ipc_callback: attrd_ipc_callback <attrd_update
> t="attrd" src="crm_attribute" task="update" attr_name="master-vmrd-res:0"
> attr_value="20000" attr_section="status" attr_host="copper2" />
> 2010-01-11 09:37:16.425602 copper2-attrd: [4649]: debug:
> attrd_local_callback: update message from crm_attribute:
> master-vmrd-res:0=20000
> 2010-01-11 09:37:16.425611 copper2-attrd: [4649]: debug: debug2:
> find_hash_entry: #011master-vmrd-res:0->section: status
> 2010-01-11 09:37:16.425642 copper2-attrd: [4649]: debug: debug2:
> log_hash_entry: Found (and updated) entry:
> 2010-01-11 09:37:16.425651 copper2-attrd: [4649]: debug: debug2:
> log_hash_entry: Set:     status
> 2010-01-11 09:37:16.425658 copper2-attrd: [4649]: debug: debug2:
> log_hash_entry: Name:    master-vmrd-res:0
> 2010-01-11 09:37:16.425667 copper2-attrd: [4649]: debug: debug2:
> log_hash_entry: Value:   20000
> 2010-01-11 09:37:16.425699 copper2-attrd: [4649]: debug: debug2:
> log_hash_entry: Timeout: (null)
> 2010-01-11 09:37:16.425708 copper2-attrd: [4649]: debug:
> attrd_local_callback: Supplied: 20000, Current: 20000, Stored: 5
> 2010-01-11 09:37:16.425740 copper2-attrd: [4649]: debug: debug2:
> attrd_local_callback: Ignoring non-change
>
>
Hi again,

Thanks to the pacemaker log, I could browse the source code for this issue.
The stored_value is only updated after cib_ok returned, but current is
updated to the supplied value on request before sending update to cib. So
this can cause the difference in current and stored value. The code compares
only the supplied and current value. So if the cib update fails, then it
ends up this situation! The fix will be very simple, either remove compares
it against the stored_value or update current only after cib_ok returns.

Thanks
hj
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20100111/59d631b6/attachment-0001.html>