A 1.5 second command
One thread, so this is dead time for every other client
1) 1) "command" 2) (integer) 1699000000 3) (integer) 250 4) (integer) 1500
Redis LATENCY Report Analyzer
Paste the output of redis-cli LATENCY LATEST. Before reading it, know this: the latency monitor is disabled by default, so an empty report means either no spikes or nothing recorded, and those look identical.
Paste below, or drop a file anywhere on this panel
Or drop a file anywhere on this panel. Nothing is uploaded: the analysis runs in this tab.
The answer appears here
Paste on the left and press Analyze. Nothing leaves this tab.
Formatted
Results
Nothing else to flag.
No formatting problems, and nothing the rules object to. Worth remembering what that covers: this reads the file you pasted, not the account or cluster it will be applied to.
No finding matches that filter.
Examples
Real input you can load into the tool above. Each one shows a different thing going wrong, because that is what the tool is for.
Nothing recorded
The report is empty, which usually means the monitor was never switched on
(empty array)
Common mistakes
These are the ones that fail silently. The config is accepted, nothing raises an error, and the consequence arrives later.
Reading an empty report as a healthy server
The monitor is disabled by default, so an empty report is also what a server with constant stalls produces.
Instead:CONFIG SET latency-monitor-threshold 100 first.
Setting the threshold at runtime only
CONFIG SET is lost on restart, so the monitoring quietly stops after the next one.
Instead:Put it in redis.conf as well.
Treating a fork event as something to tune
Fork cost scales with mapped memory and is inherent to saving. There is no setting that removes it.
Instead:Move saves to a replica, or reduce memory per instance.
Using LATENCY to find which command was slow
It names the subsystem, not the command.
Instead:Use SLOWLOG for the specific command and its arguments.
What each latency event is actually telling you
Redis names the subsystem that stalled rather than the command that suffered, which is more useful once the names are familiar.
An empty report is ambiguous, and that is the first thing to resolve
latency-monitor-threshold defaults to 0, which means the monitor records nothing at all. An empty LATENCY LATEST is therefore consistent with a perfectly healthy server and with a server that has been stalling for months. Set the threshold before concluding anything.
fork is the cost of a background save
Redis forks for BGSAVE and for AOF rewrite, and the pause scales with how much memory the process has mapped. On virtualised hardware it is often several times worse than on bare metal. Moving saves to a replica removes it from the primary entirely.
command means an O(N) call over something large
A command event is dead time for every other client, because there is one thread. It is almost always KEYS, a full HGETALL or SMEMBERS, or a wide range query. SLOWLOG names the specific command; LATENCY only says that one was slow.
expire-cycle and eviction-cycle are capacity signals
The first means a large number of keys expired at once, which is fixed by adding jitter to TTLs so they do not land in the same second. The second means the instance is at maxmemory and doing eviction work constantly, which is a sizing problem rather than a tuning one.
What this cannot see
Only events above the threshold, only since the last LATENCY RESET, and only on this node. Latency the client experienced from network or from its own connection pool never appears here at all.
Related redis tools
Redis Pipeline Calculator It removes round trips, not work Redis INFO Analyzer The numbers INFO does not print Redis SLOWLOG Analyzer What blocked everyone else Redis MEMORY STATS Analyzer Which number actually matters Redis Bigkeys Output Analyzer Elements are not bytes Redis CLIENT LIST Analyzer Find the connection hurting you
All 53 redis tools