This page lists Cloud Monitoring metrics available for Memorystore for Redis Cluster, and describes what each metric measures.
Cloud Monitoring metrics
Metric name | Description |
---|---|
redis.googleapis.com/cluster/clients/average_connected_clients |
Mean current number of client connections across the cluster. |
redis.googleapis.com/cluster/clients/maximum_connected_clients |
Maximum current number of client connections across the cluster. |
redis.googleapis.com/cluster/clients/total_connected_clients |
Current number of client connections to the cluster. |
redis.googleapis.com/cluster/stats/total_connections_received_count |
Count of cluster-level total client connections created in the last one minute. |
redis.googleapis.com/cluster/stats/cluster/stats/total_rejected_connections_count |
Number of connections rejected because of maxclients limit. |
redis.googleapis.com/cluster/commandstats/total_usec_count |
The total time consumed per command. |
redis.googleapis.com/cluster/commandstats/total_calls_count |
Total number of calls for this command in one minute. |
redis.googleapis.com/cluster/cpu/average_utilization |
Mean CPU utilization for the cluster from 0.0 to 1.0. |
redis.googleapis.com/cluster/cpu/maximum_utilization |
Maximum CPU utilization for the cluster from 0.0 to 1.0. |
redis.googleapis.com/cluster/stats/average_expired_keys |
Mean number of key expiration events for the primaries. |
redis.googleapis.com/cluster/stats/maximum_expired_keys |
Maximum number of key expiration events for the primaries. |
redis.googleapis.com/cluster/stats/total_expired_keys_count |
Total number of key expiration events for the primaries. |
redis.googleapis.com/cluster/stats/average_evicted_keys |
Mean number of evicted keys due to memory capacity for the primaries. |
redis.googleapis.com/cluster/stats/maximum_evicted_keys |
Maximum number of evicted keys due to memory capacity on primaries |
redis.googleapis.com/cluster/stats/total_evicted_keys_count |
Number of evicted keys due to memory capacity on primaries. |
redis.googleapis.com/cluster/keyspace/total_keys |
Number of keys stored in the cluster. |
redis.googleapis.com/cluster/stats/average_keyspace_hits |
Mean number of successful lookup of keys across the cluster. |
redis.googleapis.com/cluster/stats/maximum_keyspace_hits |
Maximum number of successful lookup of keys across the cluster. |
redis.googleapis.com/cluster/stats/total_keyspace_hits_count |
Number of successful lookup of keys across the cluster. |
redis.googleapis.com/cluster/stats/average_keyspace_misses |
Mean number of failed lookup of keys across the cluster. |
redis.googleapis.com/cluster/stats/maximum_keyspace_misses |
Maximum number of failed lookup of keys across the cluster. |
redis.googleapis.com/cluster/stats/total_keyspace_misses_count |
Total number of failed lookup of keys across the cluster. |
redis.googleapis.com/cluster/memory/average_utilization |
Mean memory utilization across the cluster from 0.0 to 1.0. |
redis.googleapis.com/cluster/memory/maximum_utilization |
Maximum memory utilization across the cluster from 0.0 to 1.0. |
redis.googleapis.com/cluster/memory/total_used_memory |
Total memory usage of the cluster. |
redis.googleapis.com/cluster/memory/size |
Memory size of the cluster. |
redis.googleapis.com/cluster/replication/average_ack_lag |
Mean replication lag (in seconds) of replicas across the cluster.Replication lag (in seconds) indicates how far replicas are lagging behind primaries. |
redis.googleapis.com/cluster/replication/maximum_ack_lag |
Maximum replication acknowledge lag (in seconds) of replicas across the cluster.Replication acknowledge lag (in seconds) indicates how far replication acknowledgements are lagging behind primaries. |
redis.googleapis.com/cluster/replication/average_offset_diff |
Mean replication acknowledge offset diff (in bytes) across the cluster.Replication acknowledge offset diff means the number of bytes that have not been replicated between replicas and their primaries. |
redis.googleapis.com/cluster/replication/maximum_offset_diff |
Maximum replication offset diff (in bytes) across the cluster.Replication offset diff means the number of bytes that have not been replicated between a replicas and their primaries. |
redis.googleapis.com/cluster/stats/total_net_input_bytes_count |
Count of incoming network bytes received by the cluster endpoints. |
redis.googleapis.com/cluster/stats/total_net_output_bytes_count |
Count of outgoing network bytes sent from the cluster endpoints. |
Persistence metrics
This sections lists persistence metrics and provides sample use cases for persistence metrics.
RDB persistence metrics
Metric name | Description |
---|---|
redis.googleapis.com/cluster/persistence/rdb_saves_count |
This metric shows the cumulative number of times your cluster has taken an RDB snapshot (also known as save). This metric has a status_code field. To check if a snapshot has failed, you can filter the status_code field for the following error: 3 - INTERNAL_ERROR |
redis.googleapis.com/cluster/persistence/rdb_save_ages |
This metric shows a distribution snapshot age for all nodes across the cluster. Ideally you want to see the distribution have values that have less lag time (or the same time) than your snapshot frequency. |
AOF persistence metrics
Metric name | Description |
---|---|
redis.googleapis.com/cluster/persistence/aof_fsync_lags |
This metrics shows a distribution of the lag (from data write to durable storage sync) for all nodes in the cluster. It is only emitted for clusters with appendfsync=everysec. Ideally you want to see the distribution have values that have less lag time (or the same time) than your AOF sync frequency. |
redis.googleapis.com/cluster/persistence/aof_rewrite_count |
This metric shows the cumulative number of times for your cluster that a node has triggered an AOF rewrite. This metric has a status_code field. To check if AOF rewrites are failing, you can filter the status_code field for the following error: 3 - INTERNAL_ERROR |
Sample use cases for persistence metrics
Checking if AOF write operations cause latency and memory pressure
Suppose that you detect increased latency or memory usage on your cluster. In this case you may want to check if the extra usage is related to AOF persistence.
Since you know AOF rewrite operations can trigger transient load spikes, you can inspect the aof_rewrites_count
metric which gives you the cumulative count of AOF rewrites over the lifetime of the cluster. Suppose this metric shows you that increments in the rewrites count correspond to latency increases. In this circumstance you could address the issue by reducing the write rate or increasing the shard count to reduce the frequency of rewrites.
Checking if RDB save operations cause latency and memory pressure
Suppose that you detect increased latency or memory usage on your cluster. In this case you may want to check if the extra usage is related to RDB persistence.
Since you know RDB save operations can trigger transient load spikes, you can inspect the rdb_saves_count
metric which gives the cumulative count of RDB saves over the lifetime of the cluster. Suppose this metric shows you that increments in the RDB saves count correspond to latency increases. In this circumstance you could reduce the RDB snapshot interval to lower the frequency of rewrites. You could also scale out the cluster to reduce the baseline load levels.
Interpreting metrics for Memorystore for Redis Cluster
As seen in the list above, many of the metrics share three categories: average, maximum, and total.
For Memorystore for Redis Cluster, we provide average and maximum variations of the same metric so you can use them both to identify hotspotting for that metric family.
The total value for the metric is independent, and provides separate insight unrelated to the hotspotting purpose of average and maximum.
Understanding average and maximum metrics
Suppose you compare the average_keyspace_hits
and maximum_keyspace_hits
values for your cluster. As the difference between the two metrics grows, a
greater difference indicates more hot spotting of hits in your instance. Ideally
you would have a close value between average_keyspace_hits
and
maximum_keyspace_hits
, because this means that hits are more evenly
distributed across your instance.
This principle applies to all metrics that have the average and maximum variations of the same metric.
Hot spotting example
If you compare average_keyspace_hits
and maximum_keyspace_hits
for all of
the shards in your cluster, comparing these values indicates where hot spotting
occurs. For example, suppose shards in a 6-shard cluster have the following
number of hits:
- Shard 1 – 2 hits
- Shard 2 – 2 hits
- Shard 3 – 2 hits
- Shard 4 – 2 hits
- Shard 5 – 2 hits
- Shard 6 – 8 hits
In this example the average_keyspace_hits
returns a value of 3, and the
maximum_keyspace_hits
returns 8, indicating that shard 6 is hot.