Re: MonetDB causing disk space to fill up

Hi,

On 21.07.2015, at 20:27, Anu Raj Srivastava wrote: I won't be able to share the code I am working against but I will try to create a new stand-alone application which can replicate the issue. I'll try to share it as soon as possible. OK.

Also, in your earlier comments you mentioned that disk space does fill up for a small period of time. Is that a desired behavior? What if the connection pool stays active for long period and that causes disk space to fill up? I think this could be the scenario in my case This is expected behavior. MonetDB puts intermediate results into memory-mapped files. For these, some space is required. These files are removed once the query is finished (or should be). This has nothing to do with connections.

Hannes

-----Original Message----- From: developers-list [mailto:developers-list-bounces+anu.srivastava=rms.com@monetdb.org] On Behalf Of Hannes Mühleisen Sent: Monday, July 20, 2015 11:19 PM To: Communication channel for developers of the MonetDB suite. Subject: Re: MonetDB causing disk space to fill up

Hey Anu,

we need more than a snippet of Java code to properly debug this. Schema, sample data, queries, runnable stand-alone Java source at the least.

Hannes

...

On 20.07.2015, at 23:28, Anu Raj Srivastava wrote:

I am sorry didn¹t realize mailing list is not their in my reply.

On 7/20/15, 14:24, "Ying Zhang" wrote:

...

Hello Anu,

Would you please post this to the MonetDB users mailing list? Thanks!

Regards,

Jennie

...

On Jul 20, 2015, at 19:44, Anu Raj Srivastava wrote:

Hi Hannes,

I am using apache dbcp2 connection Pool. I am running queries in a constant loop with 8 concurrent invocations. And each 8 threads are running 9 queries each again and again.

Java code is quite straight forward:-

private ConnectionPoolPG() { try { final Properties properties = new Properties();

properties.load(ConnectionPoolPG.class.getClassLoader().getResourceA sStre am("monetDB.properties")); dataSource = BasicDataSourceFactory.createDataSource(properties); schemaName = properties.getProperty("schemaname"); } catch (Exception e) { throw new ExposureStoreException("Unable to instantiate connection pool: " + e.getMessage(), EAPAException.ERR_STORE_CONNECTION_CONFIG_ERROR); } }

public void AllQueryPerformance() throws Exception { while (true) { if (queries != null) { for (String query : queries) { try (Connection connection = ConnectionPoolPG.getInstance().getConnection()) { try (Statement statement = connection.createStatement()) { statement.execute(query); } } } } } }

-----Original Message----- From: Hannes Mühleisen [mailto:Hannes.Muehleisen@cwi.nl] Sent: Sunday, July 19, 2015 4:47 AM To: Communication channel for developers of the MonetDB suite. Cc: Ying Zhang; Anu Raj Srivastava Subject: Re: MonetDB causing disk space to fill up

...

On 19 Jul 2015, at 13:23, Hannes Mühleisen wrote:

Hi,

...

On 16 Jul 2015, at 23:11, Ying Zhang wrote:

On Jul 13, 2015, at 22:32, Anu Raj Srivastava wrote: > > Hey Jennie, > > I did upgrade to latest version but I am facing the same issue. > Further debugging into the issue it seems like it is > ConnectionPool issue. When I use new connection for each query > MonetDB seems to be able to clean up the resources

Ok, MonetDB is doing what it¹s supposed to do.

> but if I use ConnectionPool for some reason DiskSpace keep > filling up.

If a connection is kept open, I¹d expect MonetDB to be less eager to clean up resources. However, we generally recommend using a ConnectionPool, to avoid too much overhead caused by making a new connection for every query. So my suggestion for now would be to keep using the ConnectionPool, but try to find the reason why your disk is filled up.

> Do you have any idea why that could happen?

Hmmm, no. We¹ve fixed quite some leaks in Jul2015... Some more thinking: - After how many iterations of repeatedly executing the queries is your disk filled up? - Can you please try to run your queries with mclient to observe the behaviour of MonetDB (and disk usage)?

@Hannes: sorry to disturb you during holiday. But if possible, would you please have a look at this thread. Does anything ring a bell? Thanks!

I tested this with ac3p0 connection pool (limited to 4 concurrent connections) and repeating queries from 10 threads (SELECT MIN(something) FROM Š). I was not able to observe any (!) increase in storage footprint. So without a minimal reproducible example of Java code + DB schema + sample data, it is hard to tell what is going on. Is this running in auto commit? Using temporary tables?

Small follow-up: When forcing the allocation of a large intermediate (SELECT MIN(something+1) FROM Š), additional storage space is indeed claimed. The attached plot shows dbfarm size over time while 10 threads are running 1000 queries each using a 5-connection pool. You can see how the amount of additional storage drops back to exactly where it was once the pool shuts down at around 350s, and that there is no accumulation of additional storage over the course of the pool being active.

Hannes

_______________________________________________ developers-list mailing list developers-list@monetdb.org https://www.monetdb.org/mailman/listinfo/developers-list

_______________________________________________ developers-list mailing list developers-list@monetdb.org https://www.monetdb.org/mailman/listinfo/developers-list _______________________________________________ developers-list mailing list developers-list@monetdb.org https://www.monetdb.org/mailman/listinfo/developers-list