banshee/services-collection/services/redis/ssh/replication

Redis replication, by SSH

Returns the replication of a Redis server, from either side, read by redis-cli INFO on the Redis server over SSH. On a master, one link per replica connected: its state and its lag. On a replica, one link, master: the state of its link to the master and the seconds since they last talked. The role of the server, and the number of links, so that a replica that is gone says so. Five watches are offered, each one set on the probe. No password leaves banshee.

Versions

Redis 2.8 and later, and Valkey. Generic: no folder of version.

How it works

The collector runs, on the server, LC_ALL=C redis-cli -p <port> INFO, as redis/ssh/memory does, the password, when there is one, going from its file on the server into REDISCLI_AUTH. It reads the section Replication:

role:master
connected_slaves:1
slave0:ip=192.0.2.21,port=6379,state=online,offset=1093,lag=0

role:slave
master_host:192.0.2.20
master_port:6379
master_link_status:up
master_last_io_seconds_ago:1

On a master, one link per line slaveN, named <ip>:<port>: state is online, or wait_bgsave, send_bulk while the replica loads its first copy; lag is the seconds since the replica last acknowledged.
On a replica, the link master: state is master_link_status, up or down; lag is master_last_io_seconds_ago, not returned while the link is down. The replicas chained behind it are returned beside it, as on a master.
links is the number of links returned. A replica that disconnects disappears from the master's list: "No replication link" raises when none is left, and "Replica missing" names the one gone, once its name is set on the probe.
role is master or slave, as Redis writes it: after a failover (Sentinel), "Role changed" raises once the expected role is set on the probe.
A Redis that refuses the connection is critical: "Redis is not running". A refused account stops the collector misconfigured, as does any other answer than INFO.
Nothing is written: INFO only reads.

Prerequisites

The same as redis/ssh/memory: an SSH account without write rights, redis-cli in its path, the password, when Redis asks for one, in a file of that account, mode 0400 (redis:password-file), no sudo, and the fingerprint of the server's host key.

Variables

Name	Label	Says	Default
`redis:port`	Port	the port Redis listens on, on the server	`6379`
`redis:password-file`	Password file	the path, on the server, of the file that holds the password; empty: no password	—

Outputs

Name	Label	Unit	Per instance
`role`	Role	—	no
`links`	Replication links	—	no
`state`	State	—	yes, per link
`lag`	Seconds since last exchange	s	yes, per link

Watches

Name	Title	Kind	Defaults	Checked by default
`no-link`	No replication link	below a threshold, on `links`	critical below 1	yes
`down`	Replication link down	expected value, on `state`	critical when anything but `online` or `up`, for 3 runs	yes
`lag`	Replication late	above a threshold, on `lag`	warning above 10 s, critical above 60 s	yes
`role`	Role changed	expected value, on `role`	no role given: set `master` or `slave` on the probe	no
`missing`	Replica missing	expected instances, on `state`	no name given: it raises nothing until names are set on the probe	no

Deploy the collector on each server: a Redis alone, without replication, has no link, and "No replication link" would raise there.

Limits

The lag is in seconds since the last exchange, not in data behind: the replication offsets are not compared.
Redis Cluster (shards and their replicas) is not read as a cluster: each node reads as a master or a replica.
Redis is reached on the server itself (127.0.0.1), without TLS, as its default user only (see redis/ssh/memory).
Not checked against a Redis: the fields are those of the documentation of INFO (redis.io).