hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From st...@duboce.net
Subject Re: Review Request: HBASE-2832: Priorities and multi-threading for MemStore flushing
Date Tue, 13 Jul 2010 15:09:43 GMT


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java, line
20
> > <http://review.hbase.org/r/301/diff/2/?file=2551#file2551line20>
> >
> >     You couldn't call it EventHandler in the end?
> 
> Jonathan Gray wrote:
>     Yup, we can.  Will be EventType/EventHandler with master changes.  Requires the rest
of the ZK cleanup to do the full changeover.

ok


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java, line
41
> > <http://review.hbase.org/r/301/diff/2/?file=2551#file2551line41>
> >
> >     Does the base class have to know of all subclasses?
> >     
> >     http://download.oracle.com/docs/cd/E17409_01/javase/6/docs/api/java/beans/EventHandler.html
and http://download.oracle.com/docs/cd/E17409_01/javase/6/docs/api/java/util/EventObject.html
are not of use here?
> 
> Jonathan Gray wrote:
>     Not sure what you're saying with java beans classes?  You want to reuse those?  These
are quite specialized.
>     
>     We went over this stuff a few times when we did that big group review of that first
master zk patch.
>     
>     Basically what we ended up doing was trying to keep the places things are tied together
(handlers, executors, types) in enums and in as few places as possible.  You add new executor
service types in the ExecutorService class, but otherwise all the declaring/connecting of
these things is done within HBaseEventHandler.

Regards the bean classes, they cannot be used here or there is nothing in them that can help
here?

Yeah, we went over this stuff in reviews and there were the same problems then.  I thought
we'd agreed to address them.  My review is just surfacing again what was issue then.

I thought we'd agreed to stuff like a handler per event type.


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java, line
52
> > <http://review.hbase.org/r/301/diff/2/?file=2551#file2551line52>
> >
> >     A Comparable Runnable?  Thats kinda odd?  Runnable is a faceless Interface..
has nothing but a run in it... how could it be comparable?  Should be Comparable<HBaseEventHandler>?
> 
> Jonathan Gray wrote:
>     I tried a few different approaches to have prioritized stuff.
>     
>     The underlying data structure expected by the java executor services are BlockingQueue<Runnable>.
 It's trivial then to make the actual queue a PriorityBlockingQueue<Runnable> which
then just requires whatever you put in there to implement Comparable<Runnable>.  In
the compareTo we know that we will only be compared to other HBaseEventHandlers, so we can
cast and do priority/FIFO comparisons.
>     
>     I'm pretty open to other approaches (I did Comparable<HBaseEventHandler> in
one attempt already) but this turned out to be the cleanest.  Now the handlers need only override
a single getPriority() method rather than also be comparables themselves.
>     
>     Being a Comparable<HBEventHandler> means we'll have to not use a java executor
service directly, or have two separate queues.  It always starts to get a bit awkward which
is why I ended up this way.

Comparable<Runnable> punts completely on type checking.  It says less than nothing about
whats going on.  We can't have an Interface that implements Comparable and Runnable and pass
that?

My sense is that the design is broke if we're doing weird stuff like this (I've not dug in
as you have -- just passing my sense).


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java, line
151
> > <http://review.hbase.org/r/301/diff/2/?file=2551#file2551line151>
> >
> >     Why only this one handled in here?  All others in subclass?
> 
> Jonathan Gray wrote:
>     hmm, not so?  this is only RS event right now.  method immediately above it, getMasterExecutorForEvent(),
is the same style and is already committed doing master open/close handlers.

ok


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java, line
198
> > <http://review.hbase.org/r/301/diff/2/?file=2551#file2551line198>
> >
> >     Can you not use enums here?  RS_FLUSH_REGION.value rather than 64?  (where value
is datamember of the enum?)
> 
> Jonathan Gray wrote:
>     case statements must use constants.  i could switch to else if?

Can't you use enums in switch statements?


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseExecutorService.java,
line 78
> > <http://review.hbase.org/r/301/diff/2/?file=2552#file2552line78>
> >
> >     As above, this passed in int should be settting a data member (whats it setting
otherwise, the enum index?)
> 
> Jonathan Gray wrote:
>     i'm not totally clear on this int stuff with the enums.  on the event types, we actually
persist them, so I understand wanting the byte/int value.  not sure here, we can just take
it out?  there is intValue/value in there.

It seems odd.  I can understand not wanting to use enum ordinals in case you want to insert
events later but should be better way of doing this...


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseExecutorService.java,
line 157
> > <http://review.hbase.org/r/301/diff/2/?file=2552#file2552line157>
> >
> >     Whats this wait for 60 seconds about?  Hoping for an interrupt?  Why hardcoded?
> 
> Jonathan Gray wrote:
>     Need to figure out what to do here.  Wait indefinitely?
>     
>     Basically, we shutdown the executor services gracefully at first (we let running
and submitted tasks finish).  Then we'll wait for a certain period of time before interrupting
the threads running.
>     
>     I felt like it was weird to wait indefinitely but this was the behavior previously.
 I guess I should set to 0?
>     
>     (For example, MemStoreFlusher has a lock that prevents interruption while a flush
is running, so it's equivalent to a graceful shutdown)
>     
>     I suppose there is a difference here which is this also includes items waiting in
the queue.

I ain't sure whats going on here exactly.  Just commenting that the 60 seconds is odd.


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/regionserver/HRegionServer.java, line
1032
> > <http://review.hbase.org/r/301/diff/2/?file=2555#file2555line1032>
> >
> >     Why ain't this FlushEventType.startExecutorService?  There are no master flushes
so why this RS stuff in here?
> 
> Jonathan Gray wrote:
>     There is separation between what runs on master side and what runs on RS side.  Different
mapings.
>     
>     This may not even be necessary anymore, there was some reason for the separation,
need to ask Karthik on this.

ok


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/regionserver/MemStoreFlusher.java, line
270
> > <http://review.hbase.org/r/301/diff/2/?file=2556#file2556line270>
> >
> >     Whats up?  We just let the DroppedSnapshotException out now?  Or do we not throw
them anymore?
> 
> Jonathan Gray wrote:
>     As per description, handling of this moved to FlushHandler.

ok


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/regionserver/handler/FlushHandler.java,
line 76
> > <http://review.hbase.org/r/301/diff/2/?file=2557#file2557line76>
> >
> >     Your nice formatting here will not come out in javadoc.
> 
> Jonathan Gray wrote:
>     throwing <pre> tags around it enough?

yes


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/regionserver/handler/FlushHandler.java,
line 178
> > <http://review.hbase.org/r/301/diff/2/?file=2557#file2557line178>
> >
> >     If 1, 2, 3, you don't need to specify?  Just use enum ordinal?
> 
> Jonathan Gray wrote:
>     yeah it does.  seemed clearer to have the ints here i think.  i have gone back and
forth several times and whether to have these ints in the enums or not, i'm not sure why i
ended up with all of them having it.
>     
>     so you think it's better i completely remove the ints?  what about in EventHandler
where we persist it sometimes?

You can persist the ordinal... Thats fine.  There is even method in enum to give you enum
if you only have ordinal.


> On 2010-07-12 21:55:02, stack wrote:
> > trunk/src/main/java/org/apache/hadoop/hbase/regionserver/handler/FlushHandler.java,
line 220
> > <http://review.hbase.org/r/301/diff/2/?file=2557#file2557line220>
> >
> >     Low and high priority do same thing?
> 
> Jonathan Gray wrote:
>     yup.  i had/have bigger plans to get rid of MemStoreFlusher so that FlushHandler
actually does the checks and stuff.  For now it's still in the MemStoreFlusher thread and
just the flushing is multi-threaded in here.
>     
>     until that happens, same thing, that's why they call the same method, just a different
log message.
>     
>     so they do the same flush but they carry a different priority so will be executed
in priority order.

k. maybe more comment so others ain't baffled as I?


- stack


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://review.hbase.org/r/301/#review377
-----------------------------------------------------------


On 2010-07-12 22:44:39, Jonathan Gray wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://review.hbase.org/r/301/
> -----------------------------------------------------------
> 
> (Updated 2010-07-12 22:44:39)
> 
> 
> Review request for hbase, stack, Karthik Ranganathan, and Kannan Muthukkaruppan.
> 
> 
> Summary
> -------
> 
> Adds support for priorities and concurrency to regionserver flushing.
> - Adds support for RS-side events/handlers/executors
> - Adds support for prioritized HBaseEventHandlers
> - Flushing now happens through FlushHandler, a new HBaseEventHandler.  There is an RS_FLUSHER
executor pool that defaults to two threads right now but is also checking a conf value.  There
is a good bit of documentation in FlushHandler.
> - Adds unit test TestFlushHandler.  There is a nicer way to detect when flushes finish
now for other tests.
> - Handling of FS errors is pushed into FlushHandler now.  The changes happening with
the master rewrite introduce a ServerStatus interface (probably a RegionStatus for rs side)
that will contain the necessary methods rather than using HRegionServer directly as is required
for now.
> - Something weird not passing in tests with multiple masters and regionservers, still
working that out.
> 
> 
> This addresses bug HBASE-2832.
>     http://issues.apache.org/jira/browse/HBASE-2832
> 
> 
> Diffs
> -----
> 
>   trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseEventHandler.java 963507

>   trunk/src/main/java/org/apache/hadoop/hbase/executor/HBaseExecutorService.java 963507

>   trunk/src/main/java/org/apache/hadoop/hbase/master/HMaster.java 963507 
>   trunk/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java 963507

>   trunk/src/main/java/org/apache/hadoop/hbase/regionserver/HRegionServer.java 963507

>   trunk/src/main/java/org/apache/hadoop/hbase/regionserver/MemStoreFlusher.java 963507

>   trunk/src/main/java/org/apache/hadoop/hbase/regionserver/handler/FlushHandler.java
PRE-CREATION 
>   trunk/src/test/java/org/apache/hadoop/hbase/regionserver/TestFlushHandler.java PRE-CREATION

> 
> Diff: http://review.hbase.org/r/301/diff
> 
> 
> Testing
> -------
> 
> Adds TestFlushHandler which passes.  Unit tests now passing.
> 
> 
> Thanks,
> 
> Jonathan
> 
>


Mime
View raw message