To amplify on this, the collections before tenuring are copying collections
and thus immune to fragmentation. CMS, however, stands for Concurrent Mark
Sweep and mark-sweep collectors are subject to fragmentation. When
fragmentation causes a failure to allocate, then you get a full copying GC.
If
On Thu, Jan 27, 2011 at 11:38 AM, Wayne wrote:
> Should we schedule a rolling restart every 24 hours?
You could do this (See J-Ds note).
You could try a different JVM (Todd tripped over this beauty recently
that would seem to have fixes added in u20 to address fragmentation
made things worse:
ht
Doing it from the master is what we aim to do, but it's a lot more
involved. I know that jgray has a few ideas on the subject.
J-D
On Thu, Jan 27, 2011 at 4:25 PM, Ted Dunning wrote:
> Why doesn't the master do this? Why not just set it up so that you can tell
> the master that the maximum numb
Why doesn't the master do this? Why not just set it up so that you can tell
the master that the maximum number of regions for the soon-to-go region
server is 0?
On Thu, Jan 27, 2011 at 3:53 PM, Jean-Daniel Cryans wrote:
> Writing this gives me an idea... I think one "easy" way we could
> achieve
Not as far as I know, unless you disabled splits from the beginning
like some ppl do.
J-D
On Thu, Jan 27, 2011 at 4:22 PM, Ted Yu wrote:
> Is there a way to disable splitting (on a particular region server) ?
>
> On Thu, Jan 27, 2011 at 4:20 PM, Jean-Daniel Cryans
> wrote:
>
>> Mmm yes for the
Is there a way to disable splitting (on a particular region server) ?
On Thu, Jan 27, 2011 at 4:20 PM, Jean-Daniel Cryans wrote:
> Mmm yes for the sake of not having a single region that moved, but it
> wouldn't be so bad... it just means that those regions will be closed
> when the RS closes.
>
Mmm yes for the sake of not having a single region that moved, but it
wouldn't be so bad... it just means that those regions will be closed
when the RS closes.
Also it's possible to have splits during that time, again it's not
dramatic as long as the script doesn't freak out because a region is
go
Should steps 1 and 2 below be exchanged ?
Regards
On Thu, Jan 27, 2011 at 3:53 PM, Jean-Daniel Cryans wrote:
> To mitigate heap fragmentation, you could consider adding more nodes
> to the cluster :)
>
> Regarding rolling restarts, currently there's one major issue:
> https://issues.apache.org/j
To mitigate heap fragmentation, you could consider adding more nodes
to the cluster :)
Regarding rolling restarts, currently there's one major issue:
https://issues.apache.org/jira/browse/HBASE-3441
How it currently works is a bit dumb, when you cleanly close a region
server it will first close a
I assumed GC was *trying* to roll. It shows the last 30min of logs with
control characters at the end.
We are not all writes. In terms of writes we can wait and the zookeeper
timeout can go way up, but we also need to support real-time reads (end user
based) and that is why the zookeeper timeout i
On Thu, Jan 27, 2011 at 6:48 AM, Wayne wrote:
> We have got .90 up and running well, but again after 24 hours of loading a
> node went down. Under it all I assume it is a GC issue, but the GC logging
> rolls every < 60 minutes so I can never see logs from 5 hours ago (working
> on getting Scribe u
I only see the bad datanode error on the one node right before zookeeper
brought it down.
On Thu, Jan 27, 2011 at 10:53 AM, Ted Yu wrote:
> About bad datanode error, I found 164 occurrences in 7-node dev cluster
> hbase 0.90 region server logs.
> In our 14 node staging cluster running hbase 0.20
About bad datanode error, I found 164 occurrences in 7-node dev cluster
hbase 0.90 region server logs.
In our 14 node staging cluster running hbase 0.20.6, I found none.
Both use cdh3b2 hadoop.
On Thu, Jan 27, 2011 at 6:48 AM, Wayne wrote:
> We have got .90 up and running well, but again after
We have got .90 up and running well, but again after 24 hours of loading a
node went down. Under it all I assume it is a GC issue, but the GC logging
rolls every < 60 minutes so I can never see logs from 5 hours ago (working
on getting Scribe up to solve that). Most of our issues are a node being
m
14 matches
Mail list logo