You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@trafficserver.apache.org by "Thomas Jackson (JIRA)" <ji...@apache.org> on 2016/04/30 21:15:12 UTC

[jira] [Commented] (TS-4403) HostDB's "stale while revalidate" doesn't span failed DNS lookups

    [ https://issues.apache.org/jira/browse/TS-4403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15265440#comment-15265440 ] 

Thomas Jackson commented on TS-4403:
------------------------------------

After looking into this issue more, it seems more obvious that this is just an error condition that this feature didn't account for. If you completely turn off your resolvers (timeouts etc.) then the records are properly served stale. But if the resolver returns a failed lookup (NXDOMAIN, SERVFAIL, etc.) then ATS clobbers the record with a error entry (which is used internally to not constantly pound the resolvers). So, it seems obvious to me that the intent here should be to serve the stale record all the way up until we get a valid DNS response saying that it has changed (or doesn't exist) and these error conditions signify a transient error-- and as such should not make hostdb drop the record. 

> HostDB's "stale while revalidate" doesn't span failed DNS lookups
> -----------------------------------------------------------------
>
>                 Key: TS-4403
>                 URL: https://issues.apache.org/jira/browse/TS-4403
>             Project: Traffic Server
>          Issue Type: Bug
>            Reporter: Thomas Jackson
>
> Right now if you set stale_while_revalidate hostdb will return the stale record while waiting on the DNS lookup. In the event that you have the DNS timeout set low (lets say 1 second) and the DNS server times out (lets say due to an outage of your resolver) then ATS will start replacing all of your stale entries (which were working) with error timeout entries (which are empty)-- which is bad.
> The stale while revalidate configuration should serve stale records until a valid response from DNS comes back-- the patch here should be relatively simple, just a bit of a nuisance to test since DNS requires port 53 ;)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)