[Herddb-dev] New Diskless Cluster Mode v2

[Enrico Olivelli]

Hi all,
I have created a PR for Diskless Cluster Mode v2
https://github.com/diennea/herddb/pull/676

See the description for the details.

This change introduces the ability to set '*' as 'replicas' for a
tablespace, meaning that every node is automatically a good replica for the
tablespace.

This makes sense in Diskless Cluster Mode v2 because in that mode only one
node works, the leader, and followers are stand-by only and do not store
additional data.

As a follow up I would like to work in two directions:
- make the cluster mode: 'traditional' vs 'diskless' configurable
per-tablespace and not per cluster, I would call it 'storageMode: file vs
bookkeeper'
- create default tablespace 'herd' with replica = '*', storageMode:
bookkeeper and maxleaderinactivitytimeout=30 seconds

with these two features the tablespace 'herd' will be always available, and
this will ease working in an HerdDB cluster, because by default all of the
queries, event the ones about metadata (systablespaces, sysnodes.....) are
by default routed to the server for 'herd' tablespace.

An alternative would be to add a new virtualtablespace, like  'metadata'
that cannot contain tables and that it is assigned to every node
(replica=*,  maxleaderinactivitytimeout=30 seconds)

I would like to get this stuff in for 0.19.0, as I am preparing for the
ApacheCon talk, and we need to be more userfriendly

Thoughts ?
Enrico
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://webmail4.mag-news.it/pipermail/herddb-dev/attachments/20200810/69e96435/attachment.html>