Low priority question: negative cache in autofs5

Low priority question: negative cache in autofs5

[Ondrej Valousek]

Probably to Ian:
I am just looking in the autofs5 (from RHEL5) source. I have an indirect
nis map:
auto.master:
/appli auto.appli
/proj auto.proj

Now, auto.proj contains a wildcard entry (* nfsserver:/share/&) but
auto.appli does not.

1. Every time I try to enter (shortly after each other)
/proj/nonexistent_directory, I got this:
Dec  5 12:10:45 ara automount[3101]: rmdir_path: lstat of
/proj/nonexistent_directory failed
... why did not the negative cache save us the expensive call to lstat?


2. Every time I try to enter (shortly after each other)
/appli/nonexistent_directory, I got this:
Dec  5 12:11:41 ara automount[3101]: lookup_mount: lookup(yp): key
"nonexistent" not found in map
... again, looks like the nis daemon was consulted but the negative
cache should take the place.

Maybe I do not understand the concept of negative caching well....
Many thanks,

Ondrej

Re: Low priority question: negative cache in autofs5

[Ian Kent]

On Fri, 2008-12-05 at 12:30 +0100, Ondrej Valousek wrote:
> Probably to Ian:
> I am just looking in the autofs5 (from RHEL5) source. I have an indirect
> nis map:
> auto.master:
> /appli auto.appli
> /proj auto.proj
> 
> Now, auto.proj contains a wildcard entry (* nfsserver:/share/&) but
> auto.appli does not.
> 
> 1. Every time I try to enter (shortly after each other)
> /proj/nonexistent_directory, I got this:
> Dec  5 12:10:45 ara automount[3101]: rmdir_path: lstat of
> /proj/nonexistent_directory failed
> ... why did not the negative cache save us the expensive call to lstat?
> 
> 
> 2. Every time I try to enter (shortly after each other)
> /appli/nonexistent_directory, I got this:
> Dec  5 12:11:41 ara automount[3101]: lookup_mount: lookup(yp): key
> "nonexistent" not found in map
> ... again, looks like the nis daemon was consulted but the negative
> cache should take the place.

Known bug with v5 handling of negative caching of non-existent keys.
I'm working on it but it is proving a little bit tricky because of
possible multiple map sources listed in nsswitch.conf (at least that's
where I'm currently at).

> 
> Maybe I do not understand the concept of negative caching well....

No .. you got it right .. I missed it.

Another thing, there are quite a few changes going into RHEL-5.3 autofs
which will make their way into RHEL-4.8 in due course. Hopefully you
will not get caught by them but it is worth logging calls for issues in
case you are seeing a known issue.

Ian

Re: Low priority question: negative cache in autofs5

[Ondrej Valousek]

Hi Ian,

Thanks for the quick reply. And what is the answer to the first question
please (regarding lstat & wildcard entry)? :-)
It is not he same case I guess....
Many thanks,

Ondrej

Ian Kent wrote:
> On Fri, 2008-12-05 at 12:30 +0100, Ondrej Valousek wrote:
>   
>> Probably to Ian:
>> I am just looking in the autofs5 (from RHEL5) source. I have an indirect
>> nis map:
>> auto.master:
>> /appli auto.appli
>> /proj auto.proj
>>
>> Now, auto.proj contains a wildcard entry (* nfsserver:/share/&) but
>> auto.appli does not.
>>
>> 1. Every time I try to enter (shortly after each other)
>> /proj/nonexistent_directory, I got this:
>> Dec  5 12:10:45 ara automount[3101]: rmdir_path: lstat of
>> /proj/nonexistent_directory failed
>> ... why did not the negative cache save us the expensive call to lstat?
>>
>>
>> 2. Every time I try to enter (shortly after each other)
>> /appli/nonexistent_directory, I got this:
>> Dec  5 12:11:41 ara automount[3101]: lookup_mount: lookup(yp): key
>> "nonexistent" not found in map
>> ... again, looks like the nis daemon was consulted but the negative
>> cache should take the place.
>>     
>
> Known bug with v5 handling of negative caching of non-existent keys.
> I'm working on it but it is proving a little bit tricky because of
> possible multiple map sources listed in nsswitch.conf (at least that's
> where I'm currently at).
>
>   
>> Maybe I do not understand the concept of negative caching well....
>>     
>
> No .. you got it right .. I missed it.
>
> Another thing, there are quite a few changes going into RHEL-5.3 autofs
> which will make their way into RHEL-4.8 in due course. Hopefully you
> will not get caught by them but it is worth logging calls for issues in
> case you are seeing a known issue.
>
> Ian
>
>   

Re: Low priority question: negative cache in autofs5

[Ian Kent]

On Fri, 2008-12-05 at 13:20 +0100, Ondrej Valousek wrote:
> Hi Ian,
> 
> Thanks for the quick reply. And what is the answer to the first question
> please (regarding lstat & wildcard entry)? :-)

The negative cache working properly will return a fail immediately so
you shouldn't see this.

But then it shouldn't happen for the first case either.
There are quite a few changes scheduled for the next update and I
haven't seen this for a while now so it's a bit hard to get exited about
it.

> It is not he same case I guess....
> Many thanks,
> 
> Ondrej
> 
> Ian Kent wrote:
> > On Fri, 2008-12-05 at 12:30 +0100, Ondrej Valousek wrote:
> >   
> >> Probably to Ian:
> >> I am just looking in the autofs5 (from RHEL5) source. I have an indirect
> >> nis map:
> >> auto.master:
> >> /appli auto.appli
> >> /proj auto.proj
> >>
> >> Now, auto.proj contains a wildcard entry (* nfsserver:/share/&) but
> >> auto.appli does not.
> >>
> >> 1. Every time I try to enter (shortly after each other)
> >> /proj/nonexistent_directory, I got this:
> >> Dec  5 12:10:45 ara automount[3101]: rmdir_path: lstat of
> >> /proj/nonexistent_directory failed
> >> ... why did not the negative cache save us the expensive call to lstat?
> >>
> >>
> >> 2. Every time I try to enter (shortly after each other)
> >> /appli/nonexistent_directory, I got this:
> >> Dec  5 12:11:41 ara automount[3101]: lookup_mount: lookup(yp): key
> >> "nonexistent" not found in map
> >> ... again, looks like the nis daemon was consulted but the negative
> >> cache should take the place.
> >>     
> >
> > Known bug with v5 handling of negative caching of non-existent keys.
> > I'm working on it but it is proving a little bit tricky because of
> > possible multiple map sources listed in nsswitch.conf (at least that's
> > where I'm currently at).
> >
> >   
> >> Maybe I do not understand the concept of negative caching well....
> >>     
> >
> > No .. you got it right .. I missed it.
> >
> > Another thing, there are quite a few changes going into RHEL-5.3 autofs
> > which will make their way into RHEL-4.8 in due course. Hopefully you
> > will not get caught by them but it is worth logging calls for issues in
> > case you are seeing a known issue.
> >
> > Ian
> >
> >   
> 

Re: Low priority question: negative cache in autofs5

[Ondrej Valousek]

> The negative cache working properly will return a fail immediately so
> you shouldn't see this.
>
> But then it shouldn't happen for the first case either.
> There are quite a few changes scheduled for the next update and I
> haven't seen this for a while now so it's a bit hard to get exited about
> it.
>
>   
Ok, I understand - in summary:
#1 problem should go [hopefully] away with 5.3
#2 problem is known bug being worked on....
Thanks!
O.

Re: Low priority question: negative cache in autofs5

[Ian Kent]

On Fri, 2008-12-05 at 14:30 +0100, Ondrej Valousek wrote:
> > The negative cache working properly will return a fail immediately so
> > you shouldn't see this.
> >
> > But then it shouldn't happen for the first case either.
> > There are quite a few changes scheduled for the next update and I
> > haven't seen this for a while now so it's a bit hard to get exited about
> > it.
> >
> >   
> Ok, I understand - in summary:
> #1 problem should go [hopefully] away with 5.3
> #2 problem is known bug being worked on....

Right, but problem #1 may be tied to problem #2, not quite sure yet.
A fix for the negative cache issue is way too late for 5.3.
We would need to get a test package to you for that, once it's done.

> Thanks!
> O.