Эта страница ещё не переведена: показана английская версия.

Elasticsearch / OpenSearch

Elasticsearch and OpenSearch: cluster status (red or yellow), unassigned shards, disks filling up node by node, and JVM heap pressure.

Network ports to open

From the DumbMonit server (or its relay agent) to the device:

Port	Protocol	Purpose
9200	TCP	REST API (HTTP or HTTPS, default)

DumbMonit reads the REST API with a user that only holds the cluster privilege monitor: it sees the health and the statistics of the nodes, and no document of any index. Both engines speak the same dialect for this, so one device type covers them.

A red cluster has at least one primary shard that is not allocated: part of the data cannot be read or written. A yellow one has lost replicas: the next node failure will lose data. The quiet failure is the disk: at 95 % (flood stage) the engine turns the indices read-only and writes are refused.

What it watches

All metrics are prefixed dumbmonit_elasticsearch_.

Metric	What	Labels
version_info	value 1	version, cluster, distribution
cluster_status	0 green, 1 yellow, 2 red
nodes, data_nodes	nodes in the cluster
active_primary_shards, active_shards, active_shards_percent	allocated shards
unassigned_shards, delayed_unassigned_shards, initializing_shards, relocating_shards	shards that are not in service yet
pending_tasks	cluster state tasks waiting
node_disk_used_percent, node_disk_free_bytes, node_disk_total_bytes	the data disk of each node	node
node_heap_used_percent, node_cpu_percent	JVM heap and CPU of each node	node

The node statistics are optional: a user that may not read them keeps the cluster health, without the per-node series.

The built-in rules that apply: Elasticsearch cluster red, Elasticsearch cluster yellow, Elasticsearch unassigned shards, Elasticsearch node disk almost full, Elasticsearch JVM heap pressure, plus Monitor unreachable.

Create a monitoring-only Elasticsearch user

1. Create a role that only holds the cluster privilege monitor, from Kibana (Stack Management, Roles) or with the API.

   curl -u elastic -X PUT "https://es.lan:9200/_security/role/dumbmonit" -H "Content-Type: application/json" -d '{"cluster":["monitor"]}'
   

2. Create the user with that role.

   curl -u elastic -X PUT "https://es.lan:9200/_security/user/dumbmonit" -H "Content-Type: application/json" -d '{"password":"choose-a-long-password","roles":["dumbmonit"]}'
   

3. OpenSearch: create a role with the cluster permission cluster_monitor in Security, Roles, and map the user to it.

4. In DumbMonit, enter the address of one node, for example es.lan, then the user and its password.

5. Elasticsearch 8 and later serve HTTPS with a generated certificate: pick HTTPS, and tick Accept an unverifiable certificate unless you installed your own.

Required permissions

• monitor (cluster privilege): Read the cluster health and node statistics
• cluster_monitor (OpenSearch): The same permission on OpenSearch

Warning

The account needs no index privilege: DumbMonit never reads a document. One node reports the whole cluster, so enter a node that is not the one you expect to fail.

Why these permissions

monitor is the cluster privilege that covers /_cluster/health and /_nodes/stats. The user has no index privilege on purpose. The status is a number so that a rule can compare it: red is above 1, yellow equals 1.

Credentials

Credential	Fields
Username / password	The user created above, sent as HTTP basic authentication.
Bearer token	A service account token, sent as a bearer token. Elasticsearch API keys use another scheme and are not supported.
No authentication	A cluster with security switched off.

Address: a host name or IP (es.lan), host:port, or a full URL. Options: protocol (HTTP by default), port (9200), certificate check, request timeout (10 s).