Restarting influxdb overwrites default shardspace config

[XANi]

If graphite plugin is active while creating DB it messes up shardspaces(v0.8.0).

My shardspace config: https://gist.github.com/XANi/b22ad3d1174bdc851281

Example: 2 node cluster; graphite plugin set up to write into "stats" database

curl -X POST "http://localhost:8086/cluster/database_configs/stat?u=root&p=root" -d @/tmp/shardspace.json
curl -X POST "http://localhost:8086/cluster/database_configs/stats?u=root&p=root" -d @/tmp/shardspace.json

results in

after I change to write to 'stat' database, restart nodes and start writing, it okay... but after restarting one node other database's shardspace get messed up too:

So it looks to me like:

• spamming writes to unexisting db at it's creation can create "default" shardspace (probably if write comes between db create and adding its shardspaces)
• on restart whatever current default shardspace is, it gets some of it's parameters overwritten

[mradamczyk]

Same happens to me even without restarting influx and without having graphite.
https://groups.google.com/forum/?utm_medium=email&utm_source=footer#!msg/influxdb/c0CHA4ehcFM/7KBEv-PIt9EJ

[XANi]

Yeah I think it is unrelated to, just much easier to repeat under load. I "fixed" it by just leaving default shardspace and adding ones I wanted on top of it

[pauldix]

If you send writes to a non-existant database, the graphite plugin will create it. If you create a database and have no shard spaces created, it will create a default one for you.

I think the solution here is simple. Create the database AND your shard spaces before writing any data to InfluxDB. Is this not an option?

[XANi]

It does not create database in 0.8, it does complain in logs that there is no database ;]

And I did that. Turned off graphite plugin, created DB, turned plugin back on. Result ? Ok at first, after restarting database this happened:

original parameters for shard creation in https://gist.github.com/XANi/b22ad3d1174bdc851281

2 database, method of doing that was a bit different:

• Both of them only had long_term and short_term shardspaces, both with target 2
• One with "default" was created while graphite plugin was running, It looks after DB creation but before shard creation it got write, so default was autocreated.
• Other looked fine.... until after restart. long_term regexp and replica count was reset after restarting innodb

Only way I managed to make it work is:

• Naming first shard default with whatever parameters, so influxdb can freely break it on restarts
• then making rest "as i wanted them"
• effectively shardspace default exists, and it is rewritten by influxdb on restart with default parameters, but all data is catched by other shardspaces

[pauldix]

I think I can reproduce this. Will be looking into it today

[pauldix]

Fixed in 0.8.2 with this commit 64eedb7