Re: [Sks-devel] Memory Leak in recon server?

[Daniel Kahn Gillmor]

On 02/01/2010 02:22 PM, Phil Pennock wrote:
> On 2010-02-01 at 11:47 +0100, Sebastian Wiesinger wrote:
>> my sks recon server is using waaay to much memory:
>>
>> lita:~# ps auxww | fgrep sks
>> 126       4317  0.0  1.1  70584 47372 ?        S     2009  35:20 
>> /usr/sbin/sks db
>> 126       4320  0.0 40.4 3459124 1629992 ?     S     2009  24:07 
>> /usr/sbin/sks recon
>>
>> This is version 1.1.1. Any ideas why it does that? I'll restart the
>> process soon, so if you have anything I should do to debug this please
>> let me know.
> 
> One of your peers is running without having started with a populated DB
> of keys?

If i'm reading the above correctly, the VSZ is 3.5GB, and the RSS is
1.6GB.  I feel like a grouchy old man saying this, but that's a *huge*
amount of RAM.

Are you suggesting that if a single peer was to, say, flush its DB and
re-connect, it could trigger this memory consumption on all/any of its
peers?  Would the memory consumption increase proportionally to the
number of peers which did this?

It doesn't seem to me like a keyserver should need 1.6GB of RAM, esp.
not for only onepeer -- is this really a requirement?  Can we improve it
somehow?  I know that running SKS will require a lot of disk space, but
i wouldn't expect to see a recon process die on a machine with only a
half-gig of RAM.  And would this mean it's not possible for an SKS
server with two new peers to run on a 32-bit platform (address space
exhausted after 4GB?)

or is this all mmap'ed address ranges somehow?

Sorry for the confusion, and that i have way more questions than
answers.  It would be nice to be able to offer some reasonable-sounding
resource requirements people who want to volunteer to run a keyserver.


Any ideas what the resource limits are (or should be?) given the current
size and state of the WoT?

        --dkg

signature.asc
Description: OpenPGP digital signature