On ParserRegistry and classloaders
by Sanne Grinovero
I see the ParserRegistry was changed quite a bit; in Infinispan 6 it
allowed to specify a different classloader for some operations, now it
only takes a classloader during construction time.
For WildFly/JBoss users, it is quite common that the configuration
file we want parsed is not in the same classloader that the
ParserRegistry needs to lookup its own parser components (as its
design uses the ServiceLoader to discover components of the parser).
This is especially true when Infinispan is not used by the application
directly but via another module (so I guess also critical for
capedwarf).
I initially though to workaround the problem using a "wrapping
classloader" so that I could pass a single CL instance which would try
both the deployment classloader and the Infinispan's module
classloader, but - besides this being suboptimal - it doesn't work as
I'm violating isolation between modules: I can get exposed to an
Infinispan 6 module which contains also Parser components, which get
loaded as a service but are not compatible (different class
definition).
I'll need these changes in the ParserRegistry reverted please. Happy
to send a pull myself, but before I attempt to patch it myself could
someone explain what the goal was?
thanks,
Sanne
11 years, 10 months
DIST-SYNC, put(), a problem and a solution
by Bela Ban
Hi guys,
sorry for the long post, but I do think I ran into an important problem
and we need to fix it ... :-)
I've spent the last couple of days running the IspnPerfTest [1] perftest
on Google Compute Engine (GCE), and I've run into a problem with
Infinispan. It is a design problem and can be mitigated by sizing thread
pools correctly, but cannot be eliminated entirely.
Symptom:
--------
IspnPerfTest has every node in a cluster perform 20'000 requests on keys
in range [1..20000].
80% of the requests are reads and 20% writes.
By default, we have 25 requester threads per node and 100 nodes in a
cluster, so a total of 2500 requester threads.
The cache used is NON-TRANSACTIONAL / dist-sync / 2 owners:
<namedCache name="clusteredCache">
<clustering mode="distribution">
<stateTransfer awaitInitialTransfer="true"/>
<hash numOwners="2"/>
<sync replTimeout="20000"/>
</clustering>
<transaction transactionMode="NON_TRANSACTIONAL"
useEagerLocking="true"
eagerLockSingleNode="true" />
<locking lockAcquisitionTimeout="5000" concurrencyLevel="1000"
isolationLevel="READ_COMMITTED" useLockStriping="false" />
</namedCache>
It has 2 owners, a lock acquisition timeout of 5s and a repl timeout of
20s. Lock stripting is off, so we have 1 lock per key.
When I run the test, I always get errors like those below:
org.infinispan.util.concurrent.TimeoutException: Unable to acquire lock
after [10 seconds] on key [19386] for requestor [Thread[invoker-3,5,main]]!
Lock held by [Thread[OOB-194,ispn-perf-test,m5.1,5,main]]
and
org.infinispan.util.concurrent.TimeoutException: Node m8.1 timed out
Investigation:
------------
When I looked at UNICAST3, I saw a lot of missing messages on the
receive side and unacked messages on the send side. This caused me to
look into the (mainly OOB) thread pools and - voila - maxed out !
I learned from Pedro that the Infinispan internal thread pool (with a
default of 32 threads) can be configured, so I increased it to 300 and
increased the OOB pools as well.
This mitigated the problem somewhat, but when I increased the requester
threads to 100, I had the same problem again. Apparently, the Infinispan
internal thread pool uses a rejection policy of "run" and thus uses the
JGroups (OOB) thread when exhausted.
I learned (from Pedro and Mircea) that GETs and PUTs work as follows in
dist-sync / 2 owners:
- GETs are sent to the primary and backup owners and the first response
received is returned to the caller. No locks are acquired, so GETs
shouldn't cause problems.
- A PUT(K) is sent to the primary owner of K
- The primary owner
(1) locks K
(2) updates the backup owner synchronously *while holding the lock*
(3) releases the lock
Hypothesis
----------
(2) above is done while holding the lock. The sync update of the backup
owner is done with the lock held to guarantee that the primary and
backup owner of K have the same values for K.
However, the sync update *inside the lock scope* slows things down (can
it also lead to deadlocks?); there's the risk that the request is
dropped due to a full incoming thread pool, or that the response is not
received because of the same, or that the locking at the backup owner
blocks for some time.
If we have many threads modifying the same key, then we have a backlog
of locking work against that key. Say we have 100 requester threads and
a 100 node cluster. This means that we have 10'000 threads accessing
keys; with 2'000 writers there's a big chance that some writers pick the
same key at the same time.
For example, if we have 100 threads accessing key K and it takes 3ms to
replicate K to the backup owner, then the last of the 100 threads waits
~300ms before it gets a chance to lock K on the primary owner and
replicate it as well.
Just a small hiccup in sending the PUT to the primary owner, sending the
modification to the backup owner, waitting for the response, or GC, and
the delay will quickly become bigger.
Verification
----------
To verify the above, I set numOwners to 1. This means that the primary
owner of K does *not* send the modification to the backup owner, it only
locks K, modifies K and unlocks K again.
I ran the IspnPerfTest again on 100 nodes, with 25 requesters, and NO
PROBLEM !
I then increased the requesters to 100, 150 and 200 and the test
completed flawlessly ! Performance was around *40'000 requests per node
per sec* on 4-core boxes !
Root cause
---------
*******************
The root cause is the sync RPC of K to the backup owner(s) of K while
the primary owner holds the lock for K.
*******************
This causes a backlog of threads waiting for the lock and that backlog
can grow to exhaust the thread pools. First the Infinispan internal
thread pool, then the JGroups OOB thread pool. The latter causes
retransmissions to get dropped, which compounds the problem...
Goal
----
The goal is to make sure that primary and backup owner(s) of K have the
same value for K.
Simply sending the modification to the backup owner(s) asynchronously
won't guarantee this, as modification messages might get processed out
of order as they're OOB !
Suggested solution
----------------
The modification RPC needs to be invoked *outside of the lock scope*:
- lock K
- modify K
- unlock K
- send modification to backup owner(s) // outside the lock scope
The primary owner puts the modification of K into a queue from where a
separate thread/task removes it. The thread then invokes the PUT(K) on
the backup owner(s).
The queue has the modified keys in FIFO order, so the modifications
arrive at the backup owner(s) in the right order.
This requires that the way GET is implemented changes slightly: instead
of invoking a GET on all owners of K, we only invoke it on the primary
owner, then the next-in-line etc.
The reason for this is that the backup owner(s) may not yet have
received the modification of K.
This is a better impl anyway (we discussed this before) becuse it
generates less traffic; in the normal case, all but 1 GET requests are
unnecessary.
Improvement
-----------
The above solution can be simplified and even made more efficient.
Re-using concepts from IRAC [2], we can simply store the modified *keys*
in the modification queue. The modification replication thread removes
the key, gets the current value and invokes a PUT/REMOVE on the backup
owner(s).
Even better: a key is only ever added *once*, so if we have [5,2,17,3],
adding key 2 is a no-op because the processing of key 2 (in second
position in the queue) will fetch the up-to-date value anyway !
Misc
----
- Could we possibly use total order to send the updates in TO ? TBD (Pedro?)
Thoughts ?
[1] https://github.com/belaban/IspnPerfTest
[2]
https://github.com/infinispan/infinispan/wiki/RAC:-Reliable-Asynchronous-...
--
Bela Ban, JGroups lead (http://www.jgroups.org)
12 years, 1 month
Cache.size() on distributed caches?
by Alan Field
Hey,
I have been looking at adding the ability to get the total size of a cache in RadarGun. The first implementation I coded used the distributed iterators in Infinispan 7.[1] I then realized that implementing getTotalSize() method using a distributed executor would allow the code in versions back to Infinispan 5.2. I have the code written, and I have been running some Jenkins jobs with Infinispan 6.0.1 Final to verify that the results are correct.[2] I use the RandomData stage to put data in the cache. Here is what it writes in the log:
04:11:59,573 INFO [org.radargun.stages.cache.RandomDataStage] (main) Received responses from all 4 slaves. Durations [0 = 17.04 minutes, 1 = 18.36 minutes, 2 = 18.44 minutes, 3 = 18.58 minutes]
04:11:59,574 INFO [org.radargun.stages.cache.RandomDataStage] (main) --------------------
04:11:59,574 INFO [org.radargun.stages.cache.RandomDataStage] (main) Filled cache with String objects totaling 25% of the Java heap
04:11:59,574 INFO [org.radargun.stages.cache.RandomDataStage] (main) Slave 0 wrote 479352 values to the cache with a total size of 958,704 kb; targetMemoryUse = 1,022,368 kb; countOfWordsInData = 44900952
04:11:59,575 INFO [org.radargun.stages.cache.RandomDataStage] (main) Slave 1 wrote 479352 values to the cache with a total size of 958,704 kb; targetMemoryUse = 1,022,368 kb; countOfWordsInData = 44914319
04:11:59,575 INFO [org.radargun.stages.cache.RandomDataStage] (main) Slave 2 wrote 479352 values to the cache with a total size of 958,704 kb; targetMemoryUse = 1,022,368 kb; countOfWordsInData = 44906729
04:11:59,575 INFO [org.radargun.stages.cache.RandomDataStage] (main) Slave 3 wrote 479352 values to the cache with a total size of 958,704 kb; targetMemoryUse = 1,022,368 kb; countOfWordsInData = 44908687
04:11:59,576 INFO [org.radargun.stages.cache.RandomDataStage] (main) The cache contains 1917408 values with a total size of 3,834,816 kb
04:11:59,576 INFO [org.radargun.stages.cache.RandomDataStage] (main) 100 words were generated with a maximum length of 20 characters
04:11:59,576 INFO [org.radargun.stages.cache.RandomDataStage] (main) --------------------
These are the outputs from my getTotalSize() code:
04:11:59,591 INFO [org.radargun.service.Infinispan53CacheInfo] (main) org.radargun.service.Infinispan53CacheInfo$Cache.getTotalSize() for cache testCache
04:12:12,094 INFO [org.radargun.service.Infinispan53CacheInfo] (main) cache.size() = 1917408
04:12:26,283 INFO [org.radargun.service.Infinispan53CacheInfo] (main) cache.getAdvancedCache().size() = 1917408
04:12:26,283 INFO [org.radargun.service.Infinispan53CacheInfo] (main) cache.getAdvancedCache().getCacheConfiguration().clustering().hash().numOwners() = 2
04:12:26,283 INFO [org.radargun.service.Infinispan53CacheInfo] (main) cache.getCacheManager().getMembers().size() = 4
04:12:41,955 INFO [org.radargun.stages.cache.ClearCacheStage] (main) Cache size = 3834800
The "Cache size =" message is from the results of my distributed executor, and the other messages are informational. These outputs show that calling cache size on a distributed cache returns the size of the entire cache including any passivated entries, not just the size of the cache on the local node. This breaks the code of my distributed executor, but mostly makes it unnecessary if I can just call cache.size().
Is this an expected change in behavior?
Thanks,
Alan
[1] https://github.com/radargun/radargun/blob/master/plugins/infinispan70/src...
[2] https://jenkins.mw.lab.eng.bos.redhat.com/hudson/job/jdg-radargun-gettota...
--
Alan Field
Principal Quality Engineer - JBoss Data Grid
T: (919) 890-8932 | Ext. 8148932
12 years, 1 month
JPA & OSGi.. still a long way to go.
by Sanne Grinovero
Hi all,
I just noticed that "ISPN-4276 - Make JPA cache store work in Karaf"
was resolved.
I trust that a single instance might work now, but we need to take
into considerations some limitations of running Hibernate in OSGi, in
particular the caveats documented here:
http://docs.jboss.org/hibernate/orm/4.2/devguide/en-US/html/ch17.html#d5e...
The classloader being overriden by an OSGi deployment, it overrides a
static which affects all instances of Hibernate running in a specified
classloader. This is meant to be fixed in Hibernate ORM 5.0.
Considering that a JPACacheStore can handle a single entity, and
consequentially we need to suggest users to use a separate Cache for
each type - but you can't give this suggestion when running in Karaf
so "work in Karaf" is still needing some love unless we intend to
resolve only the single Cache - single type use case.
This is in addition of the functional limitations (like no support for
relations) that we already discussed in a different context: in OSGi,
you can't run more than one JPACacheStore currently.
Where should these limitations be tracked?
-- Sanne
12 years, 1 month
A question and an observation
by Bela Ban
1: Observation:
-------------
In my Infinispan perf test (IspnPerfTest), I used
cache.getAdvancedCache().withFlags(...).put(key,value) in a tight loop.
I've always thought that withFlags() was a fast operation, *but this is
not the case* !!
Once I changed this and predefined the 2 caches (sync and async) at the
start, outside the loop, things got 10x faster ! So please change this
if you made the same mistake !
2. Question:
-----------
In Infinispan 6, I defined my custom transport as follows:
<transport ... transportClass="org.perf.CustomTransport"/>
This is gone in 7. Do I now have to use programmatic configuration ? If
so, how would I do this ?
--
Bela Ban, JGroups lead (http://www.jgroups.org)
12 years, 1 month
Propagate the schema to the cachestore
by Emmanuel Bernard
A remark by Divya made me think of something.
With Infinispan moving to the direction of ProtoBuf and schemas, cache store would greatly benefit from receiving in one shape or another that schema to transform a blob into something more structure depending on the underlying capability of the datastore.
Has anyone explored that angle?
Emmanuel
12 years, 1 month
