Swapping, not fragmentation
A ratio below 1.0, which is the more urgent problem and the better-looking number
total.allocated 536870912 fragmentation 0.85
Redis MEMORY STATS Analyzer
Paste the output of redis-cli MEMORY STATS. The number that matters is rarely the biggest one: a fragmentation ratio below 1.0 is more urgent than a large overhead figure, and it reads like the better number.
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.
Real fragmentation
Above 1.5 the allocator is holding memory the dataset is not using
total.allocated 1073741824 fragmentation 2.10 fragmentation.bytes 400000000
Overhead dominating
Most of the memory is bookkeeping, which means too many tiny keys
total.allocated 1000000000 overhead.total 700000000 keys.count 5000000 keys.bytes-per-key 60
Common mistakes
These are the ones that fail silently. The config is accepted, nothing raises an error, and the consequence arrives later.
Reading a fragmentation ratio below 1.0 as good
It means resident memory is smaller than allocated, so the process is partly in swap. That is worse than fragmentation, not better.
Instead:Check the operating system and remove the swap pressure.
Running activedefrag because the ratio is not 1.0
Defragmentation cannot help a ratio below 1.0, and it costs CPU. It only addresses the case above roughly 1.5.
Instead:Read which side of 1.0 the number is on first.
Sizing the machine from current allocation
Memory is released to the allocator but often not to the operating system, so resident size tracks the peak rather than the current total.
Instead:Size for peak.allocated.
Treating one node's output as the cluster's
Each node reports only its own memory. A shard with a problem does not appear in another shard's numbers.
Instead:Collect it from every node.
Four numbers that are read the wrong way round
MEMORY STATS is thirty-odd fields with no ordering by importance. These are the ones where the obvious reading is the wrong one.
A fragmentation ratio BELOW 1.0 means swapping
The ratio is resident memory divided by allocated memory. Above 1.0 the allocator is holding extra, which is fragmentation. Below 1.0 the resident set is SMALLER than what Redis allocated, which can only mean part of the process is in swap. On a single-threaded server every swapped page turns a microsecond command into a disk read, and it looks like a lower, healthier number.
Overhead is not waste, until it is most of the total
Overhead covers the main dictionary, the expires table, client buffers and the replication backlog. Some is unavoidable. When it passes half the allocated total, the instance is mostly bookkeeping, and that shape almost always means a very large number of very small keys rather than anything misconfigured.
Bytes per key includes the key name
Every key costs a dictionary entry, a redis object and the key string, roughly 50 to 100 bytes before any value. When bytes-per-key is under about 100, the overhead is comparable to the data, and grouping those keys into hashes is the single biggest saving available.
Peak is what the machine has to fit
Redis releases memory when keys go away, but the allocator does not necessarily return it to the operating system, so resident size stays near the peak and the fragmentation ratio then reports the gap. Size the machine for the peak, not for what MEMORY STATS says right now.
What this cannot see
One instance at one moment. On a cluster each node reports only its own memory, so a problem on one shard is invisible in another's output, and a quiet sample says nothing about behaviour under load.
More redis tools
Redis Hash Slot Calculator Which of the 16,384 slots does this key land in? Redis Hash Tag Tester Will these keys survive a multi-key command? Redis RESP Protocol Decoder Read what the server actually sent Redis RESP2 vs RESP3 Reply Decoder What the same reply looks like on each protocol Redis Command to RESP Exactly what your client puts on the socket Redis Glob Pattern Tester Redis globs are not shell globs Redis Connection URL Parser The path is the database number Redis Memory Unit Converter 1g and 1gb are not the same number Redis TTL Converter -1 and -2 are not durations Redis Stream ID Parser The first half is a millisecond timestamp redis.conf Validator Will Redis start with this file? Redis Production Config Linter The settings that cause incidents redis.conf to CONFIG SET Which of these can you change without a restart? Redis ACL Rule Decoder What does this user actually get? Redis ACL Validator Find the rule that does nothing Redis INFO Analyzer The numbers INFO does not print Redis SLOWLOG Analyzer What blocked everyone else Redis CLUSTER NODES Parser Read the topology, and find the gap Redis Cluster Slot Distribution Who owns how much, and what is missing Redis Keyspace Notification Flags Why your events never arrive Redis Memory Calculator The encoding decides, not the data Redis Key Count to Memory The fixed cost per key Redis Encoding Threshold Calculator One field more, several times the memory Redis Bitmap Memory Calculator Sized by the highest bit, not the set ones Redis HyperLogLog Calculator 12 KB whether you count a thousand or a billion Redis Cluster Sizing Only 60% of each node is usable Redis RDB and AOF Size Calculator The fork needs memory, not disk Redis Replication Bandwidth Calculator How long the backlog actually covers Redis Connection Pool Calculator More connections is not more throughput Redis Pipeline Calculator It removes round trips, not work Redis Cache Hit Rate Calculator 99% to 90% is ten times the backend load Redis Eviction Policy Simulator volatile- with no TTLs is noeviction Redis Cost Estimator Your rates, so nothing goes stale Redis Config File Generator A redis.conf with the reasons in it Redis ACL Generator A user that can do exactly one job Redis Maxmemory Config Generator The limit, and the headroom it needs Redis Persistence Config Generator How much you can afford to lose Redis Lua Script Generator Atomic, and short enough to stay that way Redis TLS Config Generator Encrypted, and the old port actually closed Redis Sentinel Config Generator Failover that can actually be authorised Redis Cluster Config Generator Three primaries, and the bus port open Redis Docker Compose Generator Local Redis that is not on the internet Redis Client Config Generator Timeouts on both sides, and a sane pool Redis Bigkeys Output Analyzer Elements are not bytes Redis CLIENT LIST Analyzer Find the connection hurting you Redis LATENCY Report Analyzer An empty report may mean nothing was recorded Redis Keyspace Prefix Analyzer Which key family is growing Redis SET Command Builder A plain SET clears the TTL Redis ZRANGE Query Builder REV reverses the argument order Redis SCAN Iteration Planner COUNT is a hint, not a page size Redis Key Name Validator Legal is not the same as workable Redis Cluster Compatibility Checker Works now, breaks when you shard