> On Jan. 22, 2014, 8:50 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?
>
> Bill Havanki wrote:
> 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.
do we claim to follow the TTL? I think this is just a matter of adjusting our
back off window to make sure we aren't getting stale results.
should the existing log message about sleeping specify that it's a back-off
mechanism?
something like
log.info("Backing off due to failure. Current sleep window is " + sleep /
1000. + " seconds");
would that make it more understandable as the value increases?
- Sean
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/17192/#review32532
-----------------------------------------------------------
On Jan. 22, 2014, 9:37 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, 9:37 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
>
>
