> On 22 Jun 2016, at 10:58 AM, Matt Fischer <[email protected]> wrote:
> 
> Have you setup token caching at the service level? Meaning a Memcache cluster 
> that glance, Nova etc would talk to directly? That will really cut down the 
> traffic.
> 
Yeah we have that although the default cache time is 10 seconds for revocation 
lists. I might just set that to some large number to limit this traffic a bit.

Sam



> On Jun 21, 2016 5:55 PM, "Sam Morrison" <[email protected] 
> <mailto:[email protected]>> wrote:
> 
>> On 22 Jun 2016, at 9:42 AM, Matt Fischer <[email protected] 
>> <mailto:[email protected]>> wrote:
>> 
>> On Tue, Jun 21, 2016 at 4:21 PM, Sam Morrison <[email protected] 
>> <mailto:[email protected]>> wrote:
>>> 
>>> On 22 Jun 2016, at 1:45 AM, Matt Fischer <[email protected] 
>>> <mailto:[email protected]>> wrote:
>>> 
>>> I don't have a solution for you, but I will concur that adding revocations 
>>> kills performance especially as that tree grows. I'm curious what you guys 
>>> are doing revocations on, anything other than logging out of Horizon?
>>> 
>> 
>> Is there a way to disable revocations?
>> 
>> Sam
>> 
>> 
>> I don't think so. There is no no-op driver for it that I can see. I've not 
>> tried it but maybe setting the expiration_buffer to a negative value would 
>> cause them to not be retained?
>> 
>> They expire at the rate your tokens expire (plus a buffer of 30 min by 
>> default) and under typical operation are not generated very often, so 
>> usually when you have say 10-20ish in the tree, its not too bad. It gets way 
>> worse when you have say 1000 of them. However, in our cloud anyway we just 
>> don't generate many. The only things that generate them are Horizon log outs 
>> and test suites that add and delete users and groups. If I knew we were 
>> generating anymore I'd probably setup an icinga alarm for them. When the 
>> table gets large after multiple test runs or we want to do perf tests we end 
>> up truncating the table in the DB. However that clearly is not a best 
>> security practice.
> 
> 
> How token TTLs are very low so I’d be willing to remove revocation. The bulk 
> of data going though our load balancers on API requests if you take glance 
> images out is requests to the revocation url. 
> 
> 

_______________________________________________
OpenStack-operators mailing list
[email protected]
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

Reply via email to