Fix SHOW MEASUREMENTS across the cluster

[li-ang]

fix SHOW MEASUREMENTS described by issue #4550

SHOW MEASUREMENTS maybe return ERR: json: cannot unmarshal array into Go value of type tsdb.MapperOutputJSON accross the cluster.

• CHANGELOG.md updated
• Rebased/mergable
• Tests pass
• Sign CLA

[otoolep]

@dgnorton -- can you review?

[dgnorton]

+1

[otoolep]

Thanks @li-ang

[aderumier]

Hi,

I'm testing the last nightly build with this commit apply,

I have previously the same error " cannot unmarshal array",

now I'm getting a new error message "use of closed network connection"

[li-ang]

@aderumier Can you show your test log?

[aderumier]

I don't see any special log in influxd.log. (is it possible to increase verbosity ?)

client test:

Visit https://enterprise.influxdata.com to register for updates, InfluxDB server management, and monitoring.
Connected to http://localhost:8086 version 0.9.6-nightly-3b035cc
InfluxDB shell 0.9.6-nightly-3b035cc

use collectd
Using database collectd
show measurements
ERR: write message type: use of closed network connection

(Note that it's still working fine with my others database, where I didn't have the unmarshal array error previously)

[li-ang]

Do you use whether InfluxDB cluster or InfluxDB single node? @aderumier

[alahiff]

With 0.9.6-nightly-3b035cc I also see the same error when running "show measurements". It seems to only happen when the number of replicas is less than the number of nodes in the cluster. Previously I had the "cannot unmarshal array" error in this situation.

[li-ang]

Thanks @alahiff ! I will confirm the commit ASAP! In InfluxDB official website , there is such a description as following The current version of InfluxDB is 0.9.4.1. Clustering, replication and high-availability should be considered in an alpha state. I also catch many clustering problems in the 0.9.6. I will focus on it and try to make clustering stable.

[aderumier]

I have a 3 nodes cluster,
the database "collectd" was auto-created with replica 1,
some datas has been writen,
and after that I have change the default policy to replica 3.

name duration replicaN default
default 0 3 true

I could be great to be able to have an option "replica" in [retention] configuration,

[li-ang]

Go it! Thanks @aderumier
As my opinion, changing replica is not a safe operation in current version because it can bring more problems. I suggest you create new retention policy with setting replica as 3 and then write data points.

[li-ang]

@aderumier @alahiff
Please refer to issue #4926 and pr #4927
Thanks!