GPFS Prometheus exporter

GPFS Prometheus exporter

The GPFS exporter collects metrics from the GPFS filesystem. The exporter supports the /metrics endpoint to gather GPFS metrics and metrics about the exporter.

Collectors

Collectors are enabled or disabled via --collector.<name> and --no-collector.<name> flags.

Name	Description	Default
mmgetstate	Collect state via mmgetstate	Enabled
mmpmon	Collect metrics from mmpmon using fs_io_s	Enabled
mount	Check status of GPFS mounts.	Enabled
config	Collect configs via 'mmdiag --config'	Enabled
verbs	Test if GPFS is using verbs interface	Disabled
mmhealth	Test node health through mmhealth	Disabled
waiter	Collect waiters via 'mmdiag --waiters'	Disabled
mmdf	Collect filesystem space for inodes, block and metadata.	Disabled
mmces	Collect state of CES	Disabled
mmrepquota	Collect fileset quota information	Disabled
mmlssnapshot	Collect GPFS snapshot information	Disabled
mmlsfileset	Collect GPFS fileset information	Disabled
mmlsqos	Collect GPFS I/O performance values of a file system, when you enable Quality of Service	Disabled

mount

The default behavior of the mount collector is to collect mount statuses on GPFS mounts in /proc/mounts or /etc/fstab. The --collector.mount.mounts flag can be used to adjust which mount points to check.

mmhealth

The mmhealth statuses and events collected can be filtered with the following flags that all take a regex.

• --collector.mmhealth.ignored-component - The component regex to ignore.
• --collector.mmhealth.ignored-entityname - The entity name regex to ignore.
• --collector.mmhealth.ignored-entitytype - The entity type regex to ignore.
• --collector.mmhealth.ignored-event - The event regex to ignore.

waiter

The waiter's seconds are stored in Histogram buckets defined by --collector.waiter.buckets which is a comma separated list of durations that are converted to seconds so 1s,5s,30s,1m would have buckets of []float64{1,5,30,60}.

The flag --collector.waiter.exclude defines a regular expression of waiter names to exclude.

The flag --collector.waiter.log-reason can enable logging of waiter reasons. The reason can produce very high cardinality so it is not included in metrics.

mmdf

Due to the time it can take to execute mmdf that is an executable provided that can be used to collect mmdf via cron. See gpfs_mmdf_exporter.

Flags:

• --output - This is expected to be a path collected by the Prometheus node_exporter textfile collector
• --collector.mmdf.filesystems - A comma separated list of filesystems to collect. Default is to collect all filesystems listed by mmlsfs.

mmces

The command used to collect CES states needs a specific node name. The --collector.mmces.nodename flag can be used to specify which CES node to check. The default is FQDN of those running the exporter.

mmrepquota

• --collector.mmrepquota.filesystems - A comma separated list of filesystems to collect. Default is to collect all filesystems.
• --collector.mmrepquota.quota-types - Comma seperated list of filesystem types to collect (fileset for FILESET, user for USR, group for GRP). Default is FILESET only. Ex: fileset,user collects FILESET and USR.

mmlssnapshot

• --collector.mmlssnapshot.filesystems - A comma separated list of filesystems to collect. Default is to collect all filesystems listed by mmlsfs.
• --collector.mmlssnapshot.get-size - Pass this flag to collect snapshot sizes. This operation could take a long time depending on filesystem size, consider using gpfs_mmlssnapshot_exporter instead.

The exporter gpfs_mmlssnapshot_exporter is provided to allow snapshot collection, including size (with --collector.mmlssnapshot.get-size) to be collected with cron rather than a Prometheus scrape through the normal exporter.

mmlsfileset

• --collector.mmlsfileset.filesystems - A comma separated list of filesystems to collect. Default is to collect all filesystems listed by mmlsfs.

NOTE: This collector does not collect used inodes. To get used inodes look at using the mmrepquota collector.

mmlsqos

Displays the I/O performance values of a file system, when you enable Quality of Service for I/O operations (QoS) with the mmchqos command.

Flags:

• --collector.mmlsqos.filesystems - A comma separated list of filesystems to collect. Default is to collect all filesystems listed by mmlsfs.
• --collector.mmlsqos.timeout - Count of seconds for running mmlsqos command before timeout error will be raised. Default value is 60 seconds.
• --collector.mmlsqos.seconds - Displays the I/O performance values for the previous number of seconds. The valid range of seconds is 1-999. The default value is 60 seconds.

Sudo

Ensure the user running gpfs_exporter can execute GPFS commands necessary to collect metrics. The following sudo config assumes gpfs_exporter is running as gpfs_exporter.

Defaults:gpfs_exporter !syslog
Defaults:gpfs_exporter !requiretty
# mmgetstate collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmgetstate -Y
# mmpmon collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmpmon -s -p
# config collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmdiag --config -Y
# mmhealth collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmhealth node show -Y
# verbs collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmfsadm test verbs status
# mmdf/mmlssnapshot collector if filesystems not specified
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlsfs all -Y -T
# waiter collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmdiag --waiters -Y
# mmces collector
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmces state show *
# mmdf collector, each filesystem must be listed
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmdf project -Y
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmdf scratch -Y
# mmrepquota collector, filesystems not specified
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmrepquota -j -Y -a
# mmrepquota collector, filesystems specified
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmrepquota -j -Y project scratch
# mmlssnapshot collector, each filesystem must be listed
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlssnapshot project -s all -Y
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlssnapshot ess -s all -Y
# mmlsfileset collector, each filesystem must be listed
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlsfileset project -Y
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlsfileset ess -Y
# mmlsqos collector, each filesystem must be listed
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlsqos mmfs1 -Y
gpfs_exporter ALL=(ALL) NOPASSWD:/usr/lpp/mmfs/bin/mmlsqos ess -Y

Install

Download the latest release

Add the user that will run gpfs_exporter

groupadd -r gpfs_exporter
useradd -r -d /var/lib/gpfs_exporter -s /sbin/nologin -M -g gpfs_exporter -M gpfs_exporter

Install compiled binaries after extracting tar.gz from release page.

cp /tmp/gpfs_exporter /usr/local/bin/gpfs_exporter

Add sudo rules, see sudo section

Add systemd unit file and start service. Modify the ExecStart with desired flags.

cp systemd/gpfs_exporter.service /etc/systemd/system/gpfs_exporter.service
systemctl daemon-reload
systemctl start gpfs_exporter

Build from source

To produce the gpfs_exporter, gpfs_mmdf_exporter, and gpfs_mmlssnapshot_exporter binaries:

make build

Or

go get github.com/treydock/gpfs_exporter/cmd/gpfs_exporter
go get github.com/treydock/gpfs_exporter/cmd/gpfs_mmdf_exporter
go get github.com/treydock/gpfs_exporter/cmd/gpfs_mmlssnapshot_exporter

TLS and basic auth

gpfs_exporter supports TLS and basic auth using exporter-toolkit. To use TLS and/or basic auth, users need to use --web.config.file CLI flag as follows

gpfs_exporter --web.config.file=web-config.yaml

A sample web-config.yaml file can be fetched from exporter-toolkit repository. The reference of the web-config.yaml file can be consulted in the docs.

Grafana

There is an example GPFS Performance dashboard. See the description on that dashboard for additional information on labels needed to utilize that dashboard.

Prometheus Configuration

This is an example scrape config with some metrics excluded for HPC compute nodes with label role=compute:

- job_name: gpfs
  scrape_timeout: 2m
  scrape_interval: 3m
  relabel_configs:
  - source_labels: [__address__]
    regex: "([^.]+)..*"
    replacement: "$1"
    target_label: host
  metric_relabel_configs:
  - source_labels: [__name__,role]
    regex: gpfs_(mount|health|verbs)_status;compute
    action: drop
  - source_labels: [__name__,collector,role]
    regex: gpfs_exporter_(collect_error|collector_duration_seconds);(mmhealth|mount|verbs);compute
    action: drop
  - source_labels: [__name__,role]
    regex: "^(go|process|promhttp)_.*;compute"
    action: drop
  file_sd_configs:
  - files:
    - "/etc/prometheus/file_sd_config.d/gpfs_*.yaml"

An example scrape target configuration:

- targets:
  - c0001.example.com:9303
  labels:
    host: c0001
    cluster: example
    environment: production
    role: compute