race in xenstore?

race in xenstore?

[Juergen Gross]

Hi,

I observe a strange behaviour with our BS2000 guest reading xenstore:

When the domU is started, we are reading some values from xenstore, which are
written by a backend driver on dom0. One special entry is used for signalling
validity of the other entries. The domU tries to read it until the entry is
existing. The value of this entry is always a numerical value greater 0.

In rare cases I see the following behaviour (extracted from xenstore.log):

- the domU tries to read the entry, xenstored seems not to know it:
    IN 0x61b9c0 20100423 22:25:42 READ (bs2/vmid )
    OUT 0x61b9c0 20100423 22:25:42 ERROR (ENOENT )
- on domU side, I don't see the ENOENT, but read the string "OK"!
- a little bit later (same second) the entry is written by dom0:
    IN 0x625c50 20100423 22:25:42 WRITE (/local/domain/169/bs2/vmid 2)
    OUT 0x625c50 20100423 22:25:42 WRITE (OK )

Xen version is 3.3.1

I'm not familiar with the xenstored internals, perhaps someone has an idea
what is happening here?


Juergen

-- 
Juergen Gross                 Principal Developer Operating Systems
TSP ES&S SWE OS6                       Telephone: +49 (0) 89 3222 2967
Fujitsu Technology Solutions              e-mail: juergen.gross@ts.fujitsu.com
Domagkstr. 28                           Internet: ts.fujitsu.com
D-80807 Muenchen                 Company details: ts.fujitsu.com/imprint.html

Re: race in xenstore?

[Keir Fraser]

On 27/04/2010 09:38, "Juergen Gross" <juergen.gross@ts.fujitsu.com> wrote:

> - the domU tries to read the entry, xenstored seems not to know it:
>     IN 0x61b9c0 20100423 22:25:42 READ (bs2/vmid )
>     OUT 0x61b9c0 20100423 22:25:42 ERROR (ENOENT )
> - on domU side, I don't see the ENOENT, but read the string "OK"!
> - a little bit later (same second) the entry is written by dom0:
>     IN 0x625c50 20100423 22:25:42 WRITE (/local/domain/169/bs2/vmid 2)
>     OUT 0x625c50 20100423 22:25:42 WRITE (OK )
> 
> Xen version is 3.3.1
> 
> I'm not familiar with the xenstored internals, perhaps someone has an idea
> what is happening here?

Not sure, but it would be weird for this to be a race since xenstored is
single-threaded and deals with only one request at a time, to completion.

I would focus on why xenstored logging says ENOENT, but the domU sees 'OK'.
The xenstored logic is not so hard to figure out, compared with xend for
example. ;-)

 -- Keir