accumulo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bill Havanki" <bhava...@clouderagovt.com>
Subject Re: Review Request 17192: ACCUMULO-2225 handle IllegalArgumentExceptions from Hadoop on host lookup failure.
Date Wed, 22 Jan 2014 21:39:41 GMT


> On Jan. 22, 2014, 3:50 p.m., Bill Havanki wrote:
> > src/server/src/main/java/org/apache/accumulo/server/Accumulo.java, line 229
> > <https://reviews.apache.org/r/17192/diff/1/?file=435500#file435500line229>
> >
> >     You probably want a log statement here for the final UnknownHostException, noting
that no more retries will be attempted or something to that effect.
> 
> Sean Busbey wrote:
>     oh right! I had meant to do that. at error sound fine?

Sure!


> On Jan. 22, 2014, 3:50 p.m., Bill Havanki wrote:
> > src/server/src/main/java/org/apache/accumulo/server/Accumulo.java, line 226
> > <https://reviews.apache.org/r/17192/diff/1/?file=435500#file435500line226>
> >
> >     I think the doubling of the sleep period in the last statement of the encompassing
while loop will interfere here. The first time through, sleep will be set to TTL + 1 seconds
(assuming it's set to something valid). The next time through, that sleep will have been doubled
and will be selected for the next period instead of TTL + 1 again, because the doubled value
is bigger.
> >     
> >     Also, the TTL value is a loop invariant so you could lift it out.
> 
> Sean Busbey wrote:
>     is the interference from the doubling a problem? we just need to make sure we don't
ask for a host lookup and get a cached failure. Or should the doubling not happen if we increased
sleep based on this ttl, to avoid backing off too far?
>     
>     any worry about the ttl changing between invocations by pulling it out of the loop?
It's not likely to change while running, but could.
> 
> Sean Busbey wrote:
>     Actually, not invariant because the source of the underlying UnknownHostException
could change in the loop. Acceptable cost of deduplicating the "your ttl is infinite" error
handling talked about on ACCUMULO-2224's review?

The doubling isn't really a problem, but if we claim that we follow the TTL, then as a user
it would surprise me that we'd be doubling it between checks. Documentation somewhere could
cover it.

I wouldn't care about the TTL changing between invocations.

The TTL value is invariant, but the behavior of getAddressCacheNegativeTtl isn't. This is
where that refactoring creates a problem. (Review for future reference is https://reviews.apache.org/r/17132).
I'm -0 on leaving it like it is.


- Bill


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/17192/#review32532
-----------------------------------------------------------


On Jan. 22, 2014, 4:37 p.m., Sean Busbey wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/17192/
> -----------------------------------------------------------
> 
> (Updated Jan. 22, 2014, 4:37 p.m.)
> 
> 
> Review request for accumulo and Bill Havanki.
> 
> 
> Bugs: ACCUMULO-2225
>     https://issues.apache.org/jira/browse/ACCUMULO-2225
> 
> 
> Repository: accumulo
> 
> 
> Description
> -------
> 
> ACCUMULO-2225 handle IllegalArgumentExceptions from Hadoop on host lookup failure.
>     
>     Looks for cases where we treat IOExceptions out of Hadoop specially, then attempts
to replicate for UnknownHostExceptions that have been wrapped in IllegalArgumentExceptions.
> 
> 
> Diffs
> -----
> 
>   src/core/src/main/java/org/apache/accumulo/core/client/ZooKeeperInstance.java e02c19785c373170b2b0d47266a6988adeec7e17

>   src/server/src/main/java/org/apache/accumulo/server/Accumulo.java 253962bca5f2573fdfa87b82e145938e190842ed

>   src/server/src/main/java/org/apache/accumulo/server/master/tableOps/DeleteTable.java
1c4d4ad8af30350d93f6dcfe2291463a90c835d3 
>   src/server/src/main/java/org/apache/accumulo/server/tabletserver/Compactor.java 29b8455a2dce5ecf2195d9704980e3e9749683f4

>   src/server/src/main/java/org/apache/accumulo/server/test/randomwalk/security/SecurityHelper.java
c8d1ea0872564643342387373d85b5ac87a57540 
>   src/server/src/main/java/org/apache/accumulo/server/util/TabletOperations.java 4a39050a17f84293c52f09b62932be5b957c8a91

> 
> Diff: https://reviews.apache.org/r/17192/diff/
> 
> 
> Testing
> -------
> 
> builds on both hadoop profiles. starting functional tests now in combination with 17132
> 
> 
> Thanks,
> 
> Sean Busbey
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message