News for hackers. Stuff that matters.
Re:It is done, the PostgreSQL community rocks by eo3x0 on Tuesday April 04, 2017 @11:18PM (#14039405)
What's the latest in regards to multi-master (production ready) setups in the PostgreSQL world these days? (i.e without manual or application level sharding) Last I checked, the "solutions" that were out there weren't ready for prime time.
Re:It is done, the PostgreSQL community rocks by rpedela on Wednesday April 05, 2017 @01:21AM (#14039871)
They are working hard to get logical replication in for PG 10 (next major release), but there are still a lot of open items: https://wiki.postgresql.org/wiki/PostgreSQL_10_Open_Items

I don't know if logical replication = multi-master but I know it is a major piece of the puzzle.

Re:It is done, the PostgreSQL community rocks by dgregd on Wednesday April 05, 2017 @01:51PM (#14044029)
What kind of load (number of concurrent app users) your app has so it needs multi-master db?

It is quite easy to setup powerful pg master node (32 CPU threads, fast SSDs) and many slave nodes for read-only queries. That setup can handle quite a lot of transactions.

Re:It is done, the PostgreSQL community rocks by redwood on Wednesday April 05, 2017 @08:40AM (#14041629)
I'm always curious when people ask about "multi-master" what exsctly they mean and with which tradeoffs. What use case do you envision?
Re:It is done, the PostgreSQL community rocks by user5994461 on Wednesday April 05, 2017 @09:08AM (#14041792)
It means you can write to any node.
Re:It is done, the PostgreSQL community rocks by spacemanmatt on Wednesday April 05, 2017 @10:40AM (#14042386)
That is what people think they want when they do not understand the tradeoffs involved, typically.
Re:It is done, the PostgreSQL community rocks by user5994461 on Wednesday April 05, 2017 @01:56PM (#14044076)
There can't be high availability if there is only ever one node that accepts writes.

There can't be sharding if there is only ever one node that performs writes.

Re:It is done, the PostgreSQL community rocks by redwood on Wednesday April 05, 2017 @09:05PM (#14047119)
I'm genuinely curious what you mean here when you say high availability? Do you mean that during a distributed consensus failover timeframe, it is unacceptable for writes to be queued at the client?
Re:It is done, the PostgreSQL community rocks by user5994461 on Thursday April 06, 2017 @08:16AM (#14049849)
It means that the system is still available when one node is dead.
Re:It is done, the PostgreSQL community rocks by redwood on Wednesday April 05, 2017 @09:09PM (#14047136)
When you say that there can't be sharding without more than one one node taking writes, I totally see where you're coming from. And I agree with you. Typically that means that there is a logical isolation between which data goes to which shard so this avoids conflicts too. Usually when people talk about "multimaster" I do not believe they're thinking of this particular scenario but I may be wrong. I'm not familiar but does postgres not shard?
Re:It is done, the PostgreSQL community rocks by user5994461 on Thursday April 06, 2017 @08:19AM (#14049861)
There are two cases of multi-master.

- All nodes have all data (no sharding).

- No nodes have all data (sharding).

Re:It is done, the PostgreSQL community rocks by bsg75 on Wednesday April 05, 2017 @10:52AM (#14042484)
Multi-master works wonderfully until it does not. Then you have a terrible mess of inconsistent data.
Re:It is done, the PostgreSQL community rocks by redwood on Wednesday April 05, 2017 @09:02PM (#14047100)
Okay, but in your use case, what should happen when two nodes get conflicting writes at approximately the same time?
Re:It is done, the PostgreSQL community rocks by user5994461 on Thursday April 06, 2017 @08:26AM (#14049892)
Depends on the system. Read the documentation.
Re:It is done, the PostgreSQL community rocks by redwood on Friday April 07, 2017 @08:25AM (#14059046)
Well since you were asking about this topic, I guess what I was trying to get at is what is it that you particularly were hoping to see from the community?
Re:It is done, the PostgreSQL community rocks by user5994461 on Wednesday April 05, 2017 @09:15AM (#14041820)
SQL doesn't work well in a distributed setup. Give up on SQL and it gets easier.

The most popular hack for PostgreSQL is citusdb https://www.citusdata.com/ , of course it comes with many limitations and drops half of SQL

Re:It is done, the PostgreSQL community rocks by kod on Thursday April 06, 2017 @08:28AM (#14049912)
That's FUD, citus doesn't "drop half of sql". Anything that can be resolved to a single node works exactly like normal postgres, and the main multinode things you want (e.g. aggregations) work transparently. The actual limitations are usually easy to work around.

https://docs.citusdata.com/en/v6.1/reference/sql_workarounds...

Re:It is done, the PostgreSQL community rocks by user5994461 on Saturday April 08, 2017 @08:50AM (#14066653)
I guess they improved significantly since my last work around it.