why does useradd.bbclass loop retrying its commands?

why does useradd.bbclass loop retrying its commands?

[Peter A. Bigot]

The useradd, groupadd, and groupmems commands in useradd.bbclass are 
executed in a loop with up to 10 failed attempts before they give up.  
This appears to have always been the case, as long as that file has been 
present.

Is there any reason why an initial failed attempt to execute one of 
these commands would be expected to succeed on retry, other than because 
delaying failure gives a chance for concurrently executing task to 
complete and so satisfy a dependency?

(This isn't the root cause of the pseudo autobuilder failure under 
multilib, but it did look plausible for a while.)

Peter

Re: why does useradd.bbclass loop retrying its commands?

[Richard Purdie]

On Fri, 2014-11-14 at 23:37 -0600, Peter A. Bigot wrote:
> The useradd, groupadd, and groupmems commands in useradd.bbclass are 
> executed in a loop with up to 10 failed attempts before they give up.  
> This appears to have always been the case, as long as that file has been 
> present.
> 
> Is there any reason why an initial failed attempt to execute one of 
> these commands would be expected to succeed on retry, other than because 
> delaying failure gives a chance for concurrently executing task to 
> complete and so satisfy a dependency?

Another recipe can be altering the files and holding the lock so in
theory, yes, the retries can help.

Cheers,

Richard

Re: why does useradd.bbclass loop retrying its commands?

[Peter A. Bigot]

On 11/15/2014 10:22 AM, Richard Purdie wrote:
> On Fri, 2014-11-14 at 23:37 -0600, Peter A. Bigot wrote:
>> The useradd, groupadd, and groupmems commands in useradd.bbclass are
>> executed in a loop with up to 10 failed attempts before they give up.
>> This appears to have always been the case, as long as that file has been
>> present.
>>
>> Is there any reason why an initial failed attempt to execute one of
>> these commands would be expected to succeed on retry, other than because
>> delaying failure gives a chance for concurrently executing task to
>> complete and so satisfy a dependency?
> Another recipe can be altering the files and holding the lock so in
> theory, yes, the retries can help.

Makes sense.  I do see that extrausers.bbclass has a comment explaining 
why it only passes retries=1.

Peter

Re: why does useradd.bbclass loop retrying its commands?

[Mark Hatle]

On 11/14/14, 11:37 PM, Peter A. Bigot wrote:
> The useradd, groupadd, and groupmems commands in useradd.bbclass are
> executed in a loop with up to 10 failed attempts before they give up.
> This appears to have always been the case, as long as that file has been
> present.

If multiple recipes/packages have the user, group, or shadow databases open -- 
the system will block until they are done maniuplating the file.

> Is there any reason why an initial failed attempt to execute one of
> these commands would be expected to succeed on retry, other than because
> delaying failure gives a chance for concurrently executing task to
> complete and so satisfy a dependency?

Depending on system, cores, etc.. we've seen easily it take up to 5-7 tries 
before the files unlock and are available for processing.

--Mark

> (This isn't the root cause of the pseudo autobuilder failure under
> multilib, but it did look plausible for a while.)
>
> Peter
>