Эта страница ещё не переведена: показана английская версия.

MinIO

Beta

This integration was built from the MinIO metrics documentation, not against a running server yet. It only reads. Tell us what breaks.

A MinIO object store through its Prometheus metrics: nodes and drives that went offline, raw and usable capacity, objects and buckets, the self-healing that fails, replication failures and S3 requests that end in a server error.

DumbMonit reads /minio/v2/metrics/cluster with the JWT that mc admin prometheus generate makes, sent as Authorization: Bearer, never in the URL. The token opens the metrics pages and nothing else: no S3 access, no console, no other right. If MinIO runs with MINIO_PROMETHEUS_AUTH_TYPE=public, the pages need no token. The cluster page is required; /minio/v2/metrics/node (uptime of the node asked) is optional. Metric names changed across MinIO releases (disk became drive): both are read, and a family the server does not publish simply produces no figure. The console port (9001) is not used.

Network ports to open

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

Port	Protocol	Purpose
9000	TCP	S3 API and metrics (HTTP or HTTPS, default)

What it watches

All metrics are prefixed dumbmonit_minio_.

Metric	Labels	Meaning
version_info	version, commit	Value 1.
nodes_online, nodes_offline		Nodes of the cluster.
drives_online, drives_offline, drives		Drives of the cluster.
capacity_raw_bytes, capacity_raw_free_bytes, capacity_raw_used_percent		Raw capacity.
capacity_usable_bytes, capacity_usable_free_bytes, capacity_usable_used_percent		Usable capacity, after parity.
usage_bytes, objects, buckets		Data stored, objects, buckets.
heal_objects_failed		Objects the current self-healing run could not repair.
heal_idle_seconds		Time since the last healing activity; not published until the first run ends.
heal_active		1 when healing was active in the last five minutes.
replication_failed_objects		Objects that failed to replicate (published when the server exposes it).
s3_requests_total, s3_requests_4xx_total, s3_requests_5xx_total		S3 request counters. 5xx are server errors; 4xx are client errors.
uptime_seconds		Uptime of the node answering.

The built-in rules that apply:

• MinIO node offline: for five minutes (Warning).
• MinIO drive offline: for five minutes (Advisory).
• MinIO capacity high: usable capacity above 85 % for 30 minutes (Advisory).
• MinIO heal failures: for 30 minutes (Advisory).
• MinIO replication failures: for 30 minutes (Advisory).
• MinIO S3 server errors: more than 5 % of requests in 5xx over ten minutes (Advisory).

The monitor page

The Object store panel says what is wrong in a sentence and shows nodes and drives online, capacity used and the number of objects.

Generate a Prometheus token for DumbMonit

1. With the mc client configured for your MinIO (alias myminio), generate the token of the cluster metrics.

   mc admin prometheus generate myminio cluster
   

2. Copy the value after bearer_token in the output: that is the token.

3. In DumbMonit, enter the MinIO address (for example minio.lan), paste the token as Prometheus token and choose https if MinIO serves TLS.

Required permissions

• Prometheus token — Opens the metrics pages only, no S3 or console access

Warning

The token only opens the metrics pages, but anyone holding it can read them. If MinIO runs with MINIO_PROMETHEUS_AUTH_TYPE=public, the metrics need no token: choose No authentication.

Notes

• Replication failures and 5xx counters come from the families recent MinIO releases publish; an older release may not have them.

Credentials

Credential	Fields
Prometheus token	The bearer_token printed by mc admin prometheus generate.
No authentication	For a server with MINIO_PROMETHEUS_AUTH_TYPE=public.

Address: the host name or IP of any node, host:port, or a URL. Options:

• Protocol (http): http or https.
• Port (9000): the S3 API port, not the console (9001).
• Accept an unverifiable certificate: for a server with TLS and a self-signed certificate.
• Timeout per request (15 s).

Troubleshooting

"MinIO accepted the credentials but refused … (403)": the token is wrong or was made for another server's key. Generate a new one.

"does not look like MinIO cluster metrics": the address or port is not the S3 API of a MinIO server (9000, not the console).

A node of a distributed cluster that does not answer is reported by the others: point DumbMonit at a node that stays up, and add a ping or port monitor on each node to see which one.