Re: CTE and Recursive CTE support in SQL

[piyush]

H2 has limited support for Recursive CTE.

http://h2database.com/html/advanced.html#recursive_queries

Does Ignite supports all SQL features which H2 has ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: CTE and Recursive CTE support in SQL

[piyush]

When is Recursive CTE planned ? 

Recursive CTE is super powerful.

https://www.postgresql.org/docs/current/static/queries-with.html

https://docs.microsoft.com/en-us/sql/t-sql/queries/with-common-table-expression-transact-sql?view=sql-server-2017

 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

CTE and Recursive CTE support in SQL

[piyush]

When Common Table Expression (CTE) and Recursive CTE are planned to be part
of supported SQL syntax ?

It would be super productive to have Support for CTE as we can do Graph like
Traversals and many complex queries.



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Distributed Database as best choice for persistence

[piyush]

Thanks Dmitry



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Distributed Database as best choice for persistence

[piyush]

>> What's the point of scaling persistence manually over allowing Ignite to
scale both RAM and disk layers for you?

Why there exists an official plugin for Cassandra integration then ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Distributed Database as best choice for persistence

[piyush]

>> May I ask why don't you want to go for Ignite persistence?


To operate and scale the storage layer independently and process related to
it like Compaction for example.
I intent to bundle WebServer (Netty) + Application Server (Netty) + DB
Engine (Ignite) + Cache in single deployable jar artifacts for my
application

Also to have other features like Multi DC, Hot back up etc. which are there
in other distributed KV databases. 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Distributed Database as best choice for persistence

[piyush]

cool. Does it maintain transactions ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Distributed Database as best choice for persistence

[piyush]

which is best choice of distributed persistence for Ignite if we dont want to
use native persistence ?

Riak ?  Cassandra  ? Dynamo DB ?

Has anyone tried this ? what was the experience ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Ignite on Graal VM

[piyush]

Has anybody tried Ignite on GraalVM 1.0 ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Intermittent Spikes in Response Time

[piyush]

It's hard to tell unless actual code written is known.

Although try using this JVM parameters.

-Xss256K  ;; limit threadstack size 
-XX:MaxGCPauseMillis=25  ;; limit max GC pause to 25ms

More tuning settings here

https://apacheignite.readme.io/docs/jvm-and-system-tuning
 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Sorting or ranking TextQuery?

[piyush]

Is Ranking implemented in latest version 2.4  or at least in roadmap ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Disk Persistence Performance

[piyush]

Thanks for relevant links.

Just to be clear, When you say disk latency is (20 - 500 ms) did you mean
microsecond or millisecond ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Disk Persistence Performance

[piyush]

Thanks Denis.

In my application I intend to allocated upto 25% of Data in RAM and rest
(75%) on disk.
Benchmark link given showcases data fetched from RAM.
Are there any benchmarks around how fast is disk KV store which comes
bundled as part of Ignite ?
What's the average latency value we can expect from standard SSD ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Azul Zing JVM with Apache Ignite

[piyush]

Since Azul Zing promises less than 4-5ms pause on upto 8TB of heap,  it's
okay to store data on heap and that way you can also save some efforts which
you cant when you use unsafe class with off memory. 
Please let me know if someone thinks otherwise on this.

One other advantage I see is Azul Zing JVM uses Falcon JIT which uses LLVM
to generate more optimized code and that too very fast. For in-memory data
grid, since there are lot of computations involved.

Since I intend to use Ignite as embedded DB in my Web application server it
will be helpful to rely on JVM of that caliber. That's why I am interested
in getting feedback from anyone who has used it.





--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Re: About Apache Ignite 2.4 with jdk

[piyush]

If possible dont rebuild the source, just use JDK8 instead of JDK7.
According my observation in my project JDK8 performs lot better than JDK7
when it comes to lots of computing and dealing with GC and large Heap sizes. 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Azul Zing JVM with Apache Ignite

[piyush]

Is anybody using Azul's Zing JVM with Ignite ?
How was the experience ? Does it help in some way as they claim ?

https://www.azul.com/products/zing/



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Topic based messaging

[piyush]

thanks Dmitry.
looks like i need to stick to distributed Queue as it supports replicated
backup copy.



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Topic based messaging

[piyush]

Thanks.
Can it have extra backup copies in Ignite Cluster (for HA or fault
tolerance) ?  



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Topic based messaging

[piyush]

Has anybody used this feature ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: POSIX compliant NFS storage

[piyush]

Nobody tried NFS storage with Ignite ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Graph Query Integration

[piyush]

Native Graph Db like Neo4J etc. are not distributed.
All distributed Graph DBs like DGraph (with Badger), JanusGraph (with
Cassandra/HBase) are all based on Key Value Database for a good reason that
its a Key Value structure which provides best scaling Capabilities. I would
go to the extent saying it is because Apache Ignite is based KV structure I
keen to use it in my project. KV store is most flexible and scalable IMHO.
Modern NewSQL databases like CockroachDB and TiDB are also based on KV
structure and they not definitely not same as Postgresql and MySql. 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Custom aggregate SQL function

[piyush]

https://apacheignite-sql.readme.io/docs/custom-sql-functions

We can define the custom sql function as mentioned in the doc.
Is it possible to define custom aggregate function ?
An example would be great if possible.



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Forcing indexes on heap

[piyush]

Thanks Mikael.



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Graph Query Integration

[piyush]

Apache Ignite has SQL layer using H2 parser. 
is there any library which facilitates Graph Layer over Ignite's KV store.
For example in java community we have Gremlin and Tinkerpop. 
Is the integration of such libraries exist or planned in future ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

POSIX compliant NFS storage

[piyush]

Is Apache ignite tested with any POSIX compliant Network File System Storage
?

For example, AWS has EFS:
https://docs.aws.amazon.com/efs/latest/ug/whatisefs.html

If we mount NFS disk for each node in Ignite cluster, will data have to be
replicated or we can have single copy on NFS ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Topic based messaging

[piyush]

With Regards to "Topic Based Messaging" mentioned in official doc here 
https://apacheignite.readme.io/docs/messaging

Can it be a distributed alternative to Apache Kafka ?  
Can it have extra backup copies ?
Can it use disk persistence with extra backup copies on another node just
like tables ?
 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Disk Persistence Performance

[piyush]

Thanks Stanislav.
I didnt know we could define multiple data regions and attach them to
different cache region.
Can you give a short an example for it or perhaps link for Document/blog
which does the same ?



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Disk Persistence Performance

[piyush]

Cache is based on hit or miss ratio or simple LRU
But as I mentioned in point #3 there might be services which hit multiple
times but good latency is not the requirement. I dont want cache to evict
any records when querying to such few Tables.



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: Disk Persistence Performance

[piyush]

Thanks Stanislav.

About the use case of certain tables to be Disk only is that: 
1. OLAP Report queries on Tables for which access patterns are rare or time
bound and minimum e.g. creating management report from my web application
2. Materialized view tables which are created by listening to table cache
operation triggers/events.
3. Some Tables which are accessed by my web app service for which latency
does not matter much



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Disk Persistence Performance

[piyush]

In my case, config of application  will be1. 3 Nodes with 24GB RAM and up to
1TB of disk data 2. Ignite is embedded in Java web application server 3.
Azul Zing JVM with on heap Ignite cache of 16GB4. 100mbps network speed or
better5. each node will have to serve at least 10K req/sec. each request
will several SQL statements to executeI have few queries :Curious to know
How fast is Disk KV persistence ? since Ignite iterates over all keys and
indexes to do the computation. Is Disk KV persistence is as efficient as in
other stable NoSQL database like Cassandra ?Does the number of partitions
helps in better key lookup access from Disk ?If such performance level is
not possible can we attach different KV solution like  Xodus
  , BerkeyDB Java editon etc.Can I opt
for certain tables to be Disk only and never get cached on heap regardless
of access pattern 



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/