Coordinators auto - Unable to provide advice for the current configuration

Hiya Folks,

We had to recover a cluster last night from a backup with fdbrestore. During the restore process we started with a replication factor of double and manually set 3 coordinators on the fdbcli. After recovery we moved to triple replication and attempted to set coordinators auto. Unfortunately we’re getting the following response from fdbcli.

fdb> configure auto
Unable to provide advice for the current configuration.

After digging into a bit of the code it looks like this is occurs in parseConfig in

It appears this code parses the machine readable configuration (json?) and if any one of a number attributes are missing or not what it expects it returns an empty ConfigureAutoResult(). Where all values of the ConfigureAutoResult are just set to -1 and the isValid() call returns false This results in the Unable to provide advice for the current configuration. here

We’re hoping someone can help answer the following.

1.) What’s the best way to debug what config properties that are missing or unexpected and could be causing the parseConfig to return the invalid ConfigureAutoResult.
2.) If we can’t resolve issue #1 is it safe for us to just go ahead and manually set coordinators again with the existing 3 coordinators and add 2 more so we can get to 5 coordinators and a Fault Tolerance of 2 machines.

Cluster Info
FDB Version 6.0.18

  Redundancy mode        - triple
  Storage engine         - ssd-2
  Coordinators           - 3
  Desired Proxies        - 15
  Desired Logs           - 45
  Fault Tolerance        - 1 machine
Description:	Ubuntu 16.04.6 LTS
Release:	16.04
Codename:	xenial
Linux version 4.4.0-1050-aws (buildd@lgw01-amd64-017) (gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.5) ) #59-Ubuntu SMP Tue Jan 30 19:57:10 UTC 2018

You want fdbcli> coordinators auto and not fdbcli> configure auto.

configure auto was an FDB 4.0 feature that tried to automatically set process classes and configure a correct number of proxies/logs/resolvers for you, but never actually worked super great, and was never added to the documentation as something users are suggested to use.

1 Like

oh geez, haha :man_facepalming: Blaming @rbranson he pasted this to me, can’t believe we missed it. Late night cluster restore :zzz: