HotRod: Digging further into the encoding of data
by Galder Zamarreno
Hi,
I've been looking again at the encoding of data in Hot Rod
(http://community.jboss.org/wiki/HotRodProtocol) and there's a few things
I'm not too happy about or they're not totally clear:
1. The type appears to be pretty wasteful because the majority of types
cannot combined with others, for example: You cannot have a type that is
Boolean and Long at the same time, or Double and Character. However, you
can potentially have an Arrays of Serialized. So, I propose instead
…
[View More]separating between a meta type and type.
Meta types would be: Array, Map, Primitive and Compressed. The meta type
would be encoded using bit ops, so you could combine them in diff ways,
i.e. Array of primitives. I don't think we should support combinations off
diff collections, i.e Map of Array, or Array of Maps. It would complexity
and don't forsee an immediate req for this.
Type would be: Byte, Boolean, Character, String, Date, Double, Float,
Integer, Long, Short, Serialized, StringBuilder, and Any. These would
literals from 1 to N. Note that I've added Any to separate between two
different collections. For example, if you send an Array of String, each
individual element just follows together with its size. However, if you
send an Array of Any, each individual entry must define its type.
For Maps, we've got two options: First, no type assumptions made and let
each key/value define its own type. Or allow maps meta-type definitions to
be followed by not one but two type fields. Even if the map was of mixed
types, you could have Any, Any. My preference is for the latter.
Both type and metatype would be variable length integers.
2. Serialized will be stored as byte[] internally, no attempt to
unmarshalling will be done in Hot Rod. Clients decide how they wanna
marshall this Serialized types. They just need to gives us a byte[] and
its length.
3. To clarify something that Alex mentioned in the previous encoding data
email, Arrays and Maps are followed by the number of items in Hot Rod and
not the number of bytes. In case of Arrays of Any, each individual field
gives us its size, type and the data. In case of Array of Booleans, each
individual field comes with size and data. Size might have been optional
in each field of Array of Boolean, but it simplifies deadling with Array
of Serialized, where each individual field is a byte[] of arbitrary length.
For Maps, a similar thing happens. If we have a Map of String, Boolean, we
get key/value pair like: k=[size+data]v=[size+data]. If it's a Map of Any,
Any, we get k=[type+size+data],v=[size+data]
Thoughts?
--
Galder Zamarreño
Sr. Software Engineer
Infinispan, JBoss Cache
[View Less]
15 years, 1 month
A couple of general eviction questions
by Alex Kluge
Hi,
I have a couple of general questions about eviction.
Why is it an out of band process? From http://community.jboss.org/wiki/infinispan-eviction it looks like the eviction thread is trimming the data container to its size limit. I created an interceptor and used a clock mechanism (as in clock cache) to determine when elements should be removed from the cache. It is positioned just after the mvcc locking and if there are enough keys in a node, then one is removed. The performance impact …
[View More]here is reasonable, but the clock is much faster than a strict LRU. It also removes the possibility of the OOM when you don't run the eviction thread frequently enough. The LIRS implementation also looks reasonably fast, and so could be in lined into the cache access. With a flatter cache does it make sense for eviction to be inline?
Has any thought been given to placing an explicit limit on the amount of memory that is used? You can dynamically fetch the old generation max size at startup, then set a notification on the corresponding mbean to notify you when a threshold is exceeded. As long as memory usage remains above a limit, entries can be evicted from the cache. Once it falls below the limit, the check no longer needs to be made until the notification triggers again. This is a bit course as it looks at the total memory in use, but Java doesn't have a sizeof operator. I went down this path because the general eviction path can run into issues if the maxentries is too large.
Alex
[View Less]
15 years, 1 month
PFER lock aquisitions sometimes lasting longer than 0ms
by Galder Zamarreno
As a FYI. We've detected that sometimes PFER lock acquisition succeed
after > 0ms, which is odd. We're looking into this.
------- Forwarded message -------
From: "Galder Zamarreno" <galder(a)redhat.com>
To: "Galder Zamarreno" <galder(a)redhat.com>
Cc:
Subject: Re: Fwd: Re: Random ConcurrentWriteTest.testManyUsers() failure
Date: Fri, 19 Feb 2010 09:59:46 +0100
As a fyi as well, OwnableReentrantStripedLockContainer is in use here
hence the lock impl is OwnableReentrantLock.
…
[View More]On Fri, 19 Feb 2010 09:42:55 +0100, Galder Zamarreno <galder(a)redhat.com>
wrote:
> Hi guys,
>
> As a fyi, I've been looking at a random Infinispan 2LC testsuite failure
> for a while and based on the logs below, I suspect there might be a bug
> in PFER.
>
> In 03, Thread 5 tries to acquire a lock for PFER and successfully
> acquires it in 12, 5ms after the acquisition attempt is logged. PFER is
> 0 lock timeout, so that, in theory, should have failed.
>
> I'm looking into this.
>
> In case you wanna check the logs as well, http://drop.io/ypmes4v
> contains the logs.
>
> Cheers,
>
> ------- Forwarded message -------
> From: "Galder Zamarreno" <galder(a)redhat.com>
> To: "Brian Stansberry" <brian.stansberry(a)redhat.com>
> Cc:
> Subject: Re: Random ConcurrentWriteTest.testManyUsers() failure
> Date: Thu, 18 Feb 2010 12:28:42 +0100
>
> Hi again,
>
> I added more logging. It appears that PutFromLoadValidator is not 100%
> fail proof. This is what is happening in hibernate.log attached (I'll use
> the absolute times that are shorter):
>
> Note: @ signs are followed by system identity hashcodes to differentiate
> instances.
>
> 01. 20780: Thread 5: PutFromLoadValidator: Checks whether put is valid
> and
> it's valid since it's in pending puts
> 02. 20781: Thread 5: Invokes PFER for Customer.contacts#3 with
> CollectionCacheEntry[]
> 03. 20781: Thread 5: Attempting to lock
> MarshalledValue(instance=org.hibernate.test.cache.infinispan.functional.Customer.contacts#3.
> 04. 20784: Thread 3: PutFromLoadValidator: Record recent removal
> 05. 20784: Thread 3: Within GlobalTransaction:<eq-35632>:7276, invoked
> with command RemoveCommand for Customer.contacts#3
> 06. 20784: Thread 3: Attempting to lock
> MarshalledValue(instance=org.hibernate.test.cache.infinispan.functional.Customer.contacts#3.
> 07. 20784: Thread 3: Successfully acquired lock!
> 08. 20784: Thread 3: Executes RemoveCommand on Customer.contacts#3
> 09. 20785: Thread 3: At commit time, Customer.contacts#3 is updated with
> removed=true valid=false changed=true created=false ...
> 10. 20785: Thread 3: Attempting to unlock Customer.contacts#3
> 12. 20786: Thread 5: Successfully acquired lock!
> 13. 20786: Thread 5: Executes PFER and updates Customer.contacts#3 with
> MarshalledValue(instance=CollectionCacheEntry[]; cachedHashCode=0;
> serialized=false)@c63847
> 14. 20794: Thread 3: Within GlobalTransaction:<eq-35632>:7280, calls get
> on Customer.contacts#3
> 15. 20794: Thread 3: Fetches from container and returns
> MarshalledValue(instance=CollectionCacheEntry[]; cachedHashCode=0;
> serialized=false)@c63847
>
> So, as you can see, the PFER and remove calls interleaved in such way
> that
> putFromLoad() is allowed to get in while there's a collection
> modification
> going on. Lock attempts/releases look ok. Now, can we record a
> forthcoming
> removal as a result of a collection update earlier than where is
> happening
> in the log? Maybe an option would be to instead of recording a removal of
> the collection cache, record when an entity is updated? We could do a
> pre-invalidation, similar to what happens with update timestamps after an
> entity is modified. This would reduce the chances of a situation like the
> one above happening.
>
> WDYT?
>
> On Thu, 18 Feb 2010 09:51:53 +0100, Galder Zamarreno <galder(a)redhat.com>
> wrote:
>
>> Hmmmm, looks very similar actually, if not exactly the same scenario.
>> If I look more closely in the logs, I see:
>>
>> ...
>> 12:22:12,152: Thread 3 insets new Contact#73 but JDBC connection is not
>> flushed yet!!!
>> 12:22:12,152: Thread 3 calls put on entity cache for Contact#73
>> 12:22:12,164: Thread 3 calls remove on Customer.contacts#3 // so that
>> removes collection
>> ...
>> 12:22:12,189: Thread 5 updates Customer.contacts#3 to
>> CollectionCacheEntry[] (via a PFER call)
>> ...
>>
>> Now, if read correctly, the validator should be stopping Thread 5 from
>> loading an empty collection, since the invalidation period is 10
>> seconds.
>>
>> I'll add more logging around this to see if I can see why the validator
>> is not working correctly cos I did port the validator and the tests
>> back in Sept/Oct
>>
>> Thanks for the pointer!
>>
>> On Wed, 17 Feb 2010 17:47:42 +0100, Brian Stansberry
>> <brian.stansberry(a)redhat.com> wrote:
>>
>>> Sorry; still haven't been able to really dig in on this. Have you
>>> looked at
>>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-3817 ?
>>>
>>> On 02/17/2010 06:45 AM, Galder Zamarreno wrote:
>>>> Hi Brian,
>>>>
>>>> So, I've removed the assertions from the 2nd transaction, which is the
>>>> one that checks that the contact exists after addition, or the contact
>>>> does not exist after the removal. However, the test does still
>>>> randomly
>>>> fail. Let me explain:
>>>>
>>>> 12:22:12,098: Thread 3 removed Contact#64 from Customer#3
>>>> successfully.
>>>> Customer#3 has no Contacts
>>>> 12:22:12,144: Thread 3 starts process to add new Contact to Customer#3
>>>> 12:22:12,152: Thread 3 insets new Contact#73 but JDBC connection is
>>>> not
>>>> flushed yet!!!
>>>> 12:22:12,157: Thread 5 starts reading all Customer's 1st contact
>>>> 12:22:12,160: Thread 5 checks the Customer.contacts#3 collection cache
>>>> and finds nothing.
>>>> 12:22:12,160: Thread 5 checks Contacts table in database and finds
>>>> nothing
>>>> 12:22:12,165: Thread 3 flushes JDBC connection and Contact#73 is now
>>>> seen by the rest
>>>> 12:22:12,189: Thread 5 updates Customer.contacts#3 to
>>>> CollectionCacheEntry[] (via a PFER call)
>>>> 12:22:12,196: Thread 3 uses the Session to get Customer#3's contact
>>>> but
>>>> gets nothing, since CollectionCacheEntry[] is returned by the
>>>> collection
>>>> cache.
>>>> 12:22:12,262: Thread 3 tries to remove Customer#3's contact, but since
>>>> there's none, it fails.
>>>>
>>>> The culprit would appear to me to be line in 12:22:12,189. If it
>>>> doesn't
>>>> find any contacts, it initialises to CollectionCacheEntry[], which is
>>>> different to not storing anything at all. If nothing had been stored,
>>>> 12:22:12,196 wouldn't have returned empty but instead would have gone
>>>> to
>>>> the database and retrieved it from there. However, we, in the 2LC, we
>>>> don't control what's stored, this comes from whatever Hibernate passes
>>>> us, via CollectionLoadContext. Not sure how this can be fixed...
>>>> Anything that u can think of?
>>>>
>>>> Cheers,
>>>>
>>>> On Thu, 28 Jan 2010 18:34:24 +0100, Brian Stansberry
>>>> <brian.stansberry(a)redhat.com> wrote:
>>>>
>>>>> Have fun on holidays! Better ping me about this when you get back.
>>>>>
>>>>> On 01/28/2010 10:01 AM, Galder Zamarreno wrote:
>>>>>>
>>>>>>
>>>>>> On 01/28/2010 03:46 PM, Brian Stansberry wrote:
>>>>>>> Quick response; I'll look more later: if that scenario is what is
>>>>>>> happening, the test isn't correct. There's no reason Thread 2 *in a
>>>>>>> second tx* should have any expectations.
>>>>>>
>>>>>> Thanks for the quick reply. There's no rush at all on this. Remember
>>>>>> that I'll be away on PTO next week. Back on the 8th.
>>>>>>
>>>>>> FYI, the expectations are checked in the contactExists() method that
>>>>>> gets called after removing.
>>>>>>
>>>>>>>
>>>>>>> On 01/28/2010 06:46 AM, Galder Zamarreno wrote:
>>>>>>>> Hi Brian,
>>>>>>>>
>>>>>>>> I've been looking at a random failure of
>>>>>>>> ConcurrentWriteTest.testManyUsers() in the Infinispan cache
>>>>>>>> provider and
>>>>>>>> I wanted to pass my suspicions through you to see what you think.
>>>>>>>> There's no immediate rush on this.
>>>>>>>>
>>>>>>>> Some notes:
>>>>>>>> Customer.contacts#2 refers to Customer 2's contacts collection
>>>>>>>> cache.
>>>>>>>>
>>>>>>>> Sometimes the test can fail because of this:
>>>>>>>>
>>>>>>>> 1. 12:21:40,434: Thread 1 calls readEveryonesFirstContact
>>>>>>>> 2. 12:21:40,434: Thread 1 starts tx1
>>>>>>>> 3. 12:21:40,454: Thread 1 reads Contact#302 from Customer#2 from
>>>>>>>> db
>>>>>>>> 4. 12:21:40,483: Thread 2 calls removeContact for Customer#2
>>>>>>>> 5. 12:21:40,483: Thread 2 starts tx2
>>>>>>>> 6. 12:21:40,485: Thread 2 deletes Contact#302 from Customer#2 in
>>>>>>>> db.
>>>>>>>> 4. 12:21:40,491: Thread 2 acquires lock for updating
>>>>>>>> Customer.contacts#2
>>>>>>>> collection cache.
>>>>>>>> 5. 12:21:40,492: Thread 2 deletes Contact#302 from
>>>>>>>> Customer.contacts#2
>>>>>>>> collection cache.
>>>>>>>> 6. 12:21:40,492: Thread 2 releases lock
>>>>>>>> 7. 12:21:40,492: Thread 2 commits tx2
>>>>>>>> 7. 12:21:40,493: Thread 1 acquires lock for updating
>>>>>>>> Customer.contacts#2
>>>>>>>> 8. 12:21:40,493: Thread 1 updates Customer.contacts#2 collection
>>>>>>>> cache
>>>>>>>> with Contact#302.
>>>>>>>> 9. 12:21:40,493: Thread 1 commits tx1.
>>>>>>>> 10. 12:21:40,499: Thread 2 starts a tx and checks whether any
>>>>>>>> Contacts
>>>>>>>> exist for Customer#2.
>>>>>>>> 10. 12:21:40,542: Thread 2 fails because it expects not to be any
>>>>>>>> Contacts but finds one in the cache that Thread 1 had put.
>>>>>>>>
>>>>>>>> It looks to me that this is a failure of the test itself. The fact
>>>>>>>> that
>>>>>>>> the remove and check is done in different transactions allows
>>>>>>>> another
>>>>>>>> thread to come in the middle and load some data into the cache.
>>>>>>>>
>>>>>>>> WDYT? Do you think I'm reading this correctly?
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>>
>>>>>>>> p.s. I've attached the full log in case you wanna inspect it.
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
>
>
--
Galder Zamarreño
Sr. Software Engineer
Infinispan, JBoss Cache
[View Less]
15 years, 1 month
Hot Rod - pt3
by Galder Zamarreno
Hi all,
Re: http://community.jboss.org/wiki/HotRodProtocol
First, I've made the corresponding changes based on the feedback I got
from pt2. This included reducing the response header since clients are
already aware of what they sent, addition of topology view id to to non
dumb requests and further specification in responses when topology
changes have happened...etc.
I've also added flags to the request header that allow sending
Infinispan flags like: skip cache store, zero lock …
[View More]acquisition
timeout...etc.
Note that I've noted this as being N * 1 byte where each byte represents
a flag. However, I think this could maybe be sent more efficiently by
using XOR, i.e.
0x00 -> no flag
0x01 (0000 0001) -> zero lock acquisition
0x02 (0000 0010) -> cache mode local
0x03 (0000 0011) -> zero lock acquisition + cache mode local
0x04 (0000 0100) -> skip locking
...etc.
With 2 bytes, we could implement 16 Flags, we currently 11. However, we
could use vint as well, making sure that the most significant bit does
not mean anything flag wise. Iow, with vint, in 1 byte we'd be able to
define 7 diff flags. Thoughts?
I've also added the quit command that disconnects clients.
Finally, as far as I'm concerned, the specification is complete. I'm
leaving the quiet commands out of this initial scope (see
http://code.google.com/p/memcached/wiki/MemcacheBinaryProtocol).
Remember that quiet commands could be used so that the server buffers
responses and only when you send a non-quiet command, the server replies
with all the pending answers.
As you can see at the bottom of the wiki, I've added a local only put
request/response example so that readers get an idea of what a full
command looks like. I had received some feedback from readers saying
that it was difficult to understand how it all fit together.
I'll probably add a couple more examples for non-so-dumb and clever
request/responses but I'll held them until we have a final round of
feedback and people can indicate whether they want any other examples
appearing in the wiki.
Cheers,
--
Galder Zamarreño
Sr. Software Engineer
Infinispan, JBoss Cache
[View Less]
15 years, 1 month
CloudCacheStore Bug
by Philippe Van Dyck
Hi all,
there is a bug in CloudCacheStore that makes me feel like I am the only one
using it ;-)
in CR4 : if you change the "-" sign to "A" in getBucketName ... you need to
do the opposite somewhere (or call it every time) ;-)
WDYT ?
p
15 years, 1 month
Re: [infinispan-dev] [hibernate-dev] Querying Infinispan -- Not fetching results
by Manik Surtani
Hi Vidya
Please post on the Infinispan user forum rather than the dev mail list.
Cheers
Manik
On 15 Feb 2010, at 13:19, Vidya Sethuraman wrote:
> Hi,
>
> I am able to load data in cache and display. When I try to query, I am
> getting an empty list.
> Following is the code snippet. Please let me know what I am doing wrong.
>
> ---------------------TestRecord.java----------------
> @ProvidedId(bridge = @FieldBridge(impl = StringBridge.class))
> @Indexed(index = "…
[View More]testRecord")
> public class TestRecord implements Serializable {
>
> // @Field(store = Store.YES)
> private Date today = new java.util.Date();
> // @Field(store = Store.YES)
> private Timestamp time = new Timestamp(today.getTime());
> @Field(store = Store.YES)
> private String type;
> @Field(store = Store.YES)
> private String eventType;
> @Field(store = Store.YES)
> private int loggingLevel;
> @Field(store = Store.YES)
> private String message;
> // @Field(store = Store.YES)
> private String logTime = time.toString();
>
> ...// getters and setters
> }
>
> -----TestQuery.java----------
> cache = cacheMgr.getCache();
> QueryHelper qh = new QueryHelper(cache, new Properties(), TestRecord.class);
> QueryFactory qf = new QueryFactory(cache, qh);
> CacheQuery cq = qf.getBasicQuery("message", searchText);
> auditLogRecordList = cq.list();
> int hits = cq.getResultSize();
> log.info("hits= "+ hits);
> --------------------------------------------
>
> The hits is appearing empty.
>
> Regards,
> Vidya
> _______________________________________________
> hibernate-dev mailing list
> hibernate-dev(a)lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hibernate-dev
--
Manik Surtani
manik(a)jboss.org
Lead, Infinispan
Lead, JBoss Cache
http://www.infinispan.org
http://www.jbosscache.org
[View Less]
15 years, 1 month
Many artifacts missing in repo for CR4
by Sanne Grinovero
Hello,
all of the following modules are missing in the Maven2 repository for CR4;
some are also missing CR3:
infinispan-cachestore-jdbm
infinispan-cachestore-jdbc
infinispan-cachestore-s3
infinispan-directory-demo
infinispan-ec2-demo
infinispan-gui-demo
infinispan-jopr-plugin
infinispan-lucene-demo
infinispan-server-rest
infinispan-tools
(http://repository.jboss.com/maven2/org/infinispan/)
This wasn't intentional?
Cheers,
Sanne
15 years, 1 month
