multipath.conf and polling_interval option seems misleading

multipath.conf and polling_interval option seems misleading

[shane bradley]

After reviewing the code and doing some testing I have noticed that 
polling_interval did not work as expected.
I had reviewed the description of the option for multipath.conf and it 
conflicted with the results that I had got
testing device-mapper-multipath on RHEL4/RHEL5.

$ cat /usr/share/doc/device-mapper-multipath-0.4.7/multipath.conf.annotated
#       # name    : polling_interval
#       # scope   : multipathd
#       # desc    : interval between two path checks in seconds
#       # default : 5
#       #
#       polling_interval 10

---------

The behaviour that I had expected based on the option's description above:
check path 1
wait polling_interval
check path 2
wait polling_interval
check path 1
wait polling_interval
check path 2
wait polling_interval

However after testing the results that I got was(with multipathd -v4):
example:
check path 1
check path 2
wait polling_interval
check path 1
check path 2
wait polling_interval

---------

The behaviour I seen in RHEL4 and RHEL5 was working as design after
reviewing the code and talking to a couple engineers.

The problem it seems is how I was reading the description of the option.
 From my results in testing and talking with some engineers the 
"polling_interval" option actually means:

"The interval between checking all possible paths for all multipath paths"
----------

1) Is my assumption correct that "polling_interval" actually means:
"The interval between checking all possible paths for all multipath paths"

2) What is a better way to describe the "polling_interval" option?

3) Shouldn't we make it clearer for people who don't that that much 
experience with multipathing?

--sbradley

Re: multipath.conf and polling_interval option seems misleading

[Chandra Seetharaman]

On Fri, 2009-01-16 at 11:03 -0500, shane bradley wrote:
> After reviewing the code and doing some testing I have noticed that 
> polling_interval did not work as expected.
> I had reviewed the description of the option for multipath.conf and it 
> conflicted with the results that I had got
> testing device-mapper-multipath on RHEL4/RHEL5.
> 
> $ cat /usr/share/doc/device-mapper-multipath-0.4.7/multipath.conf.annotated
> #       # name    : polling_interval
> #       # scope   : multipathd
> #       # desc    : interval between two path checks in seconds
> #       # default : 5
> #       #
> #       polling_interval 10
> 
> ---------
> 
> The behaviour that I had expected based on the option's description above:
> check path 1
> wait polling_interval
> check path 2
> wait polling_interval
> check path 1
> wait polling_interval
> check path 2
> wait polling_interval
> 
> However after testing the results that I got was(with multipathd -v4):
> example:
> check path 1
> check path 2
> wait polling_interval
> check path 1
> check path 2
> wait polling_interval
> 
> ---------
> 
> The behaviour I seen in RHEL4 and RHEL5 was working as design after
> reviewing the code and talking to a couple engineers.
> 
> The problem it seems is how I was reading the description of the option.
>  From my results in testing and talking with some engineers the 
> "polling_interval" option actually means:
> 
> "The interval between checking all possible paths for all multipath paths"
> ----------
> 
> 1) Is my assumption correct that "polling_interval" actually means:
> "The interval between checking all possible paths for all multipath paths"
> 
> 2) What is a better way to describe the "polling_interval" option?
> 
> 3) Shouldn't we make it clearer for people who don't that that much 
> experience with multipathing?

IMO, the behavior seen is the proper behavior.

If it does as per your interpretation, the seconds between checking of
the same path will depend on the number of paths to a storage, which may
not be acceptable.

May be the wording in multipath.conf.annotated should be made clear.

> --sbradley
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel

Re: multipath.conf and polling_interval option seems misleading

[Bryn M. Reeves]

Chandra Seetharaman wrote:
>> The behaviour I seen in RHEL4 and RHEL5 was working as design after
>> reviewing the code and talking to a couple engineers.
>>
>> The problem it seems is how I was reading the description of the option.
>>  From my results in testing and talking with some engineers the 
>> "polling_interval" option actually means:
>>
>> "The interval between checking all possible paths for all multipath paths"
>> ----------
>>
>> 1) Is my assumption correct that "polling_interval" actually means:
>> "The interval between checking all possible paths for all multipath paths"
>>
>> 2) What is a better way to describe the "polling_interval" option?
>>
>> 3) Shouldn't we make it clearer for people who don't that that much 
>> experience with multipathing?
> 
> IMO, the behavior seen is the proper behavior.
> 
> If it does as per your interpretation, the seconds between checking of
> the same path will depend on the number of paths to a storage, which may
> not be acceptable.
> 
> May be the wording in multipath.conf.annotated should be made clear.

Agreed - I'd not even really thought about this until Shane brought it 
up as I'd read the description to mean exactly what the code does.

It does seem that some users are puzzled by this behaviour - can't 
hurt to make the wording a little more explicit for this option.

Regards,
Bryn.

Re: multipath.conf and polling_interval option seems misleading

[shane bradley]

Yeah the wording I think is the problem. Someone familiar with source 
code or multipathing will read that option differently.

When someone reads the word path, they think of a single path to lun, 
then wait, check the next path to single lun.
They are  not thinking of all paths are checked at same time.

After thinking about it, this would not be good behaviour because it 
could take forever (minutes) before a path is found dead.
I think the problem is just the description of the option 
"polling_interval", it should be made clearer.

--sbradley

Chandra Seetharaman wrote:
> On Fri, 2009-01-16 at 11:03 -0500, shane bradley wrote:
>   
>> After reviewing the code and doing some testing I have noticed that 
>> polling_interval did not work as expected.
>> I had reviewed the description of the option for multipath.conf and it 
>> conflicted with the results that I had got
>> testing device-mapper-multipath on RHEL4/RHEL5.
>>
>> $ cat /usr/share/doc/device-mapper-multipath-0.4.7/multipath.conf.annotated
>> #       # name    : polling_interval
>> #       # scope   : multipathd
>> #       # desc    : interval between two path checks in seconds
>> #       # default : 5
>> #       #
>> #       polling_interval 10
>>
>> ---------
>>
>> The behaviour that I had expected based on the option's description above:
>> check path 1
>> wait polling_interval
>> check path 2
>> wait polling_interval
>> check path 1
>> wait polling_interval
>> check path 2
>> wait polling_interval
>>
>> However after testing the results that I got was(with multipathd -v4):
>> example:
>> check path 1
>> check path 2
>> wait polling_interval
>> check path 1
>> check path 2
>> wait polling_interval
>>
>> ---------
>>
>> The behaviour I seen in RHEL4 and RHEL5 was working as design after
>> reviewing the code and talking to a couple engineers.
>>
>> The problem it seems is how I was reading the description of the option.
>>  From my results in testing and talking with some engineers the 
>> "polling_interval" option actually means:
>>
>> "The interval between checking all possible paths for all multipath paths"
>> ----------
>>
>> 1) Is my assumption correct that "polling_interval" actually means:
>> "The interval between checking all possible paths for all multipath paths"
>>
>> 2) What is a better way to describe the "polling_interval" option?
>>
>> 3) Shouldn't we make it clearer for people who don't that that much 
>> experience with multipathing?
>>     
>
> IMO, the behavior seen is the proper behavior.
>
> If it does as per your interpretation, the seconds between checking of
> the same path will depend on the number of paths to a storage, which may
> not be acceptable.
>
> May be the wording in multipath.conf.annotated should be made clear.
>
>   
>> --sbradley
>>
>> --
>> dm-devel mailing list
>> dm-devel@redhat.com
>> https://www.redhat.com/mailman/listinfo/dm-devel
>>     
>
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel
>   

Re: multipath.conf and polling_interval option seems misleading

[Chandra Seetharaman]

On Fri, 2009-01-16 at 16:13 +0000, Bryn M. Reeves wrote:
> Chandra Seetharaman wrote:
<snip>
> > 
> > May be the wording in multipath.conf.annotated should be made clear.
> 
> Agreed - I'd not even really thought about this until Shane brought it 
> up as I'd read the description to mean exactly what the code does.

me too. I couldn't read it otherwise. 

But, i suggested document change as the user is getting confused.
> 
> It does seem that some users are puzzled by this behaviour - can't 
> hurt to make the wording a little more explicit for this option.
> 
> Regards,
> Bryn.
> 

Re: multipath.conf and polling_interval option seems misleading

[shane bradley]

[-- Attachment #1: Type: text/plain, Size: 3341 bytes --]

Patch attached for updated wording of documentation on polling_interval.


shane bradley wrote:
> Yeah the wording I think is the problem. Someone familiar with source 
> code or multipathing will read that option differently.
>
> When someone reads the word path, they think of a single path to lun, 
> then wait, check the next path to single lun.
> They are  not thinking of all paths are checked at same time.
>
> After thinking about it, this would not be good behaviour because it 
> could take forever (minutes) before a path is found dead.
> I think the problem is just the description of the option 
> "polling_interval", it should be made clearer.
>
> --sbradley
>
> Chandra Seetharaman wrote:
>> On Fri, 2009-01-16 at 11:03 -0500, shane bradley wrote:
>>  
>>> After reviewing the code and doing some testing I have noticed that 
>>> polling_interval did not work as expected.
>>> I had reviewed the description of the option for multipath.conf and 
>>> it conflicted with the results that I had got
>>> testing device-mapper-multipath on RHEL4/RHEL5.
>>>
>>> $ cat 
>>> /usr/share/doc/device-mapper-multipath-0.4.7/multipath.conf.annotated
>>> #       # name    : polling_interval
>>> #       # scope   : multipathd
>>> #       # desc    : interval between two path checks in seconds
>>> #       # default : 5
>>> #       #
>>> #       polling_interval 10
>>>
>>> ---------
>>>
>>> The behaviour that I had expected based on the option's description 
>>> above:
>>> check path 1
>>> wait polling_interval
>>> check path 2
>>> wait polling_interval
>>> check path 1
>>> wait polling_interval
>>> check path 2
>>> wait polling_interval
>>>
>>> However after testing the results that I got was(with multipathd -v4):
>>> example:
>>> check path 1
>>> check path 2
>>> wait polling_interval
>>> check path 1
>>> check path 2
>>> wait polling_interval
>>>
>>> ---------
>>>
>>> The behaviour I seen in RHEL4 and RHEL5 was working as design after
>>> reviewing the code and talking to a couple engineers.
>>>
>>> The problem it seems is how I was reading the description of the 
>>> option.
>>>  From my results in testing and talking with some engineers the 
>>> "polling_interval" option actually means:
>>>
>>> "The interval between checking all possible paths for all multipath 
>>> paths"
>>> ----------
>>>
>>> 1) Is my assumption correct that "polling_interval" actually means:
>>> "The interval between checking all possible paths for all multipath 
>>> paths"
>>>
>>> 2) What is a better way to describe the "polling_interval" option?
>>>
>>> 3) Shouldn't we make it clearer for people who don't that that much 
>>> experience with multipathing?
>>>     
>>
>> IMO, the behavior seen is the proper behavior.
>>
>> If it does as per your interpretation, the seconds between checking of
>> the same path will depend on the number of paths to a storage, which may
>> not be acceptable.
>>
>> May be the wording in multipath.conf.annotated should be made clear.
>>
>>  
>>> --sbradley
>>>
>>> -- 
>>> dm-devel mailing list
>>> dm-devel@redhat.com
>>> https://www.redhat.com/mailman/listinfo/dm-devel
>>>     
>>
>> -- 
>> dm-devel mailing list
>> dm-devel@redhat.com
>> https://www.redhat.com/mailman/listinfo/dm-devel
>>   
>
> -- 
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel


[-- Attachment #2: multipath.conf.annotated-polling_interval.patch --]
[-- Type: text/plain, Size: 440 bytes --]

--- multipath.conf.annotated.org	2009-01-29 11:03:22.000000000 -0500
+++ multipath.conf.annotated	2009-01-29 11:16:29.000000000 -0500
@@ -18,7 +18,8 @@
 #	#
 #	# name    : polling_interval
 #	# scope   : multipathd
-#	# desc    : interval between two path checks in seconds
+#	# desc    : the interval in seconds between polling all 
+#               devices that compose up all multipath devices
 #	# values  : n > 0
 #	# default : 5
 #	#

[-- Attachment #3: Type: text/plain, Size: 0 bytes --]

Re: multipath.conf and polling_interval option seems misleading

[Bryn M. Reeves]

shane bradley wrote:
> Patch attached for updated wording of documentation on polling_interval.

Looks sane to me.

Cheers,
Bryn.