Modified code for ODE Clustering

[sudharma subasinghe <su...@cse.mrt.ac.lk>]

Hi Sathwik,

I modified the code and tested it with two nodes cluster. Deploying and
undeploying processes works fine except few things.

1 ) Some times the poller is stuck due to the lock.

2 ) After using hazelcast 3.5 it does not take existing hazelcast.xml

Can you help me to solve these things.

Thank you.

Re: Modified code for ODE Clustering

[sudharma subasinghe <su...@cse.mrt.ac.lk>]

Hi Sathwik,

DeploymentPoller and DeploymentService use putIfAbsent() before calling
lock() and tryLock(). I didn't get " If a thread not holding a lock on the
non-existent key tries to introduce the key while a lock exists on the
non-existent key,". As always it checks for putIfAbsent() how does a lock
exists on the non-existent key? As in the documentation putIfAbsent() has
same functionality as

if (!map.containsKey(key))
       return map.put(key, value);
   else
       return map.get(key);



On 23 June 2015 at 13:12, Sathwik B P <sa...@gmail.com> wrote:

> Sudharma,
>
> 1) At what point is DeploymentPoller blocking? Can you enable debug logs on
> hazelcast and watch.
>
> As per the documentation,
>
> http://docs.hazelcast.org/docs/latest/javadoc/com/hazelcast/core/IMap.html#lock%28K%29
>
> "You get a lock whether the value is present in the map or not. Other
> threads (possibly on other systems) would block on their invoke of lock()
> until the non-existent key is unlocked. If the lock holder introduces the
> key to the map, the put() operation is not blocked. If a thread not holding
> a lock on the non-existent key tries to introduce the key while a lock
> exists on the non-existent key, the put() operation blocks until it is
> unlocked. "
>
> If indeed the scenario after your investigation is same as per the
> documentation, then in DeploymentPoller first go for the tryLock and if
> lock is true and then use putIfAbsent.
>
> 2) If there are issues with 3.5, revert back to earlier version.
>
> regards,
> sathwik
>
> On Tue, Jun 23, 2015 at 10:14 AM, sudharma subasinghe <
> suba.11@cse.mrt.ac.lk
> > wrote:
>
> > Hi Sathwik,
> >
> > I modified the code and tested it with two nodes cluster. Deploying and
> > undeploying processes works fine except few things.
> >
> > 1 ) Some times the poller is stuck due to the lock.
> >
> > 2 ) After using hazelcast 3.5 it does not take existing hazelcast.xml
> >
> > Can you help me to solve these things.
> >
> > Thank you.
> >
>

Re: Modified code for ODE Clustering

[Sathwik B P <sa...@gmail.com>]

Sudharma,

1) At what point is DeploymentPoller blocking? Can you enable debug logs on
hazelcast and watch.

As per the documentation,
http://docs.hazelcast.org/docs/latest/javadoc/com/hazelcast/core/IMap.html#lock%28K%29

"You get a lock whether the value is present in the map or not. Other
threads (possibly on other systems) would block on their invoke of lock()
until the non-existent key is unlocked. If the lock holder introduces the
key to the map, the put() operation is not blocked. If a thread not holding
a lock on the non-existent key tries to introduce the key while a lock
exists on the non-existent key, the put() operation blocks until it is
unlocked. "

If indeed the scenario after your investigation is same as per the
documentation, then in DeploymentPoller first go for the tryLock and if
lock is true and then use putIfAbsent.

2) If there are issues with 3.5, revert back to earlier version.

regards,
sathwik

On Tue, Jun 23, 2015 at 10:14 AM, sudharma subasinghe <suba.11@cse.mrt.ac.lk
> wrote:

> Hi Sathwik,
>
> I modified the code and tested it with two nodes cluster. Deploying and
> undeploying processes works fine except few things.
>
> 1 ) Some times the poller is stuck due to the lock.
>
> 2 ) After using hazelcast 3.5 it does not take existing hazelcast.xml
>
> Can you help me to solve these things.
>
> Thank you.
>