[DRBD-user] Zombie, Zombie?

Andreas Semt as at computer-leipzig.de
Mon Apr 26 12:31:32 CEST 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Lars Ellenberg wrote:

> please reply to the list.
> 
> / 2004-04-26 12:17:52 +0200
> \ Andreas Semt:
> 
>>>>I use DRBD 0.6.12 (with heartbeat 1.21). Sometimes I get some real high 
>>>>load on my machine (load average around 6) and a zombie process like 
>>>>drbd_syncer_2 or drbd_syncer_3. I believe the zombie process causes the 
>>>>high load. Also i lost a drbd connection without reason (one connection 
>>>>of four).
>>>>Additional information: the zombie processes disappear after some time 
>>>>and the load is normal again. Zombies exist only on the drbd "PRIMARY" 
>>>>machine (all drbd devices in primary mode).
>>>>
>>>>My questions:
>>>>
>>>>1. Could it be a zombie process who killed the drbd connection?
>>>
>>>
>>>No. It is only killed *after* the connection was lost.
>>>Unfortunately it is not always reaped immediately.
>>>Zombie processes cannot cause load, because they are dead.
>>>They only waste some memory and process slots.
>>>
>>
>>Okay, the high load is explained in the FAQ (sorry).
>>Can you say what exactly drbd_syncer and drbd_asender doing?
> 
> 
> 
> syncer reads in the blocks that need to be synced from local disk,
> and sends over to the peer.
> 
> asender sends ACKknowledgement packets for written data, and for
> drbd-pings, and itself sends drbd-ping-requests if neccessary.
> 
> 
>>>>2. Why these Zombies live on my machine?
>>>
>>>If you care, you can use 0.6.12 CVS HEAD, which differs basically only
>>>by a "reparent to init" call right after our thread startup, as it
>>>should have been from the very begining. now, the "zombies" are reaped
>>>by init almost immediately after their death.
>>>
>>
>>Is that a common behavior of drbd (to create "zombies") or a bug or 
>>something else?
> 
> 
> Not a bug, as long as they get reaped eventually.
> Unusual bahaviour, if it takes "long" to reap them.
> A bug if they stay around for ever.
> 
> Unusual in general, because typically kernel threads should be
> reparented to init anyways.
> "historically" drbd wanted to reap its child threads itself.
> 

With every mail i learn a lot! Thanks!

-- 
Best regards,
Andreas Semt



More information about the drbd-user mailing list