Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Zookeeper, mail # user - adding a separate thread to detect network timeouts faster


Copy link to this message
-
Re: adding a separate thread to detect network timeouts faster
Michi Mutsuzaki 2013-09-11, 20:36
Slow disk does affect client <-> server ping requests since ping
requests go through the commit processor.

Here is how the current client <-> server ping request works. Say the
session timeout is set to 30 seconds.

1. The client sends a ping request if the session has been inactive
for 10 seconds (1/3 of the session timeout).
2. The client waits for ping response for another 10 seconds (1/3 of
the session timeout).
3. If the client doesn't receive ping response after 10 seconds, it
tries to connect to another server.

So in this case, it can take up to 20 seconds for the client to detect
a server failure. I think this 1/3 value is picked somewhat
arbitrarily. Maybe you can make this configurable for faster failure
detection instead of introducing another heartbeat mechanism?

--Michi
On Tue, Sep 10, 2013 at 11:32 PM, Jeremy Stribling <[EMAIL PROTECTED]> wrote:
> Hi Germán,
>
> A very quick scan of that JIRA makes me think you're talking about
> server->server heartbeats, and not client->server heartbeats (which is what
> I'm talking about).  I have not tested it explicitly or inspected that part
> of the code, but I've hit many cases in testing and production where client
> session expirations coincide with long fsync times as logged by the server.
>
> Jeremy
>
>
> On 09/10/2013 10:40 PM, German Blanco wrote:
>>
>> Hello Jeremy and all,
>>
>> my idea was that the current implementation of ping handling already does
>> not wait on disk IO.
>> I am even working in a JIRA case that is related with this:
>> https://issues.apache.org/jira/browse/ZOOKEEPER-87
>> And I have also made some tests that seem to confirm that ping handling is
>> done in a different thread than transaction handling.
>> But actually, I don't have any confirmation from any person in this
>> project. Are you sure that ping handling waits on IO for anything? Have
>> you
>> tested it?
>>
>> Regards,
>> Germán Blanco.
>>
>>
>>
>> On Tue, Sep 10, 2013 at 11:05 PM, Jeremy Stribling <[EMAIL PROTECTED]>
>> wrote:
>>
>>> Good suggestion, thanks.  At the very least, I think what we have in mind
>>> would be off by default, so users could only turn it on if they know they
>>> have relatively few clients and slow disks.  An adaptive scheme would be
>>> even better, obviously.
>>>
>>>
>>> On 09/10/2013 02:04 PM, Ted Dunning wrote:
>>>
>>>> Perhaps you should be suggesting a design that is adaptive rather than
>>>> configured and guarantees low overhead at the cost of notification time
>>>> in
>>>> extreme scenarios.
>>>>
>>>> For instance, the server can send no more than 1000 (or whatever number)
>>>> HB's per second and never more than one per second to any client.  This
>>>> caps the cost nicely.
>>>>
>>>>
>>>>
>>>> On Tue, Sep 10, 2013 at 1:59 PM, Ted Dunning
>>>> <[EMAIL PROTECTED]<mailto:
>>>> [EMAIL PROTECTED]>**> wrote:
>>>>
>>>>
>>>>      Since you are talking about client connection failure detection,
>>>>      no, I don't think that there is a major barrier other than
>>>>      actually implementing a reliable check.
>>>>
>>>>      Keep in mind the cost.  There are ZK installs with 100,000
>>>>      clients.  If these are heartbeating every 2 seconds, you have
>>>>      50,000 packets per second hitting the quorum or 10,000 per server
>>>>      if all connections are well balanced.
>>>>
>>>>      If you only have 10 clients, the network burden is nominal.
>>>>
>>>>
>>>>
>>>>      On Tue, Sep 10, 2013 at 1:34 PM, Jeremy Stribling
>>>>      <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>> wrote:
>>>>
>>>>          I mostly agree, but let's assume that a ~5x speedup in
>>>>          detecting those types of failures is considered significant
>>>>          for some people. Are there technical reasons that would
>>>>          prevent this idea from working?
>>>>
>>>>          On 09/10/2013 01:31 PM, Ted Dunning wrote:
>>>>
>>>>              I don't see the strong value here.  A few failures would
>>>>              be detected more
>>>>              quickly, but I am not convinced that this would actually