banshee/hardware-collection/hardware/redfish/https/storage

Storage of a server, by Redfish

Returns, for each storage controller, drive and volume a server's management card serves by Redfish, its health; for a drive, whether it predicts its own failure and, for an SSD, its remaining life. Four watches are offered, each one set on the probe.

Versions

Generic: written for the Storage, Drive and Volume schemas of DMTF Redfish, whose fields read here have kept their form since they appeared (2016–2017). Any management card that serves them: Dell iDRAC, HPE iLO, Supermicro, Lenovo XCC, OpenBMC. Hence the common level.

How it works

The collector reads, the account in Basic, GET https://<host:address>/redfish/v1/Systems, each system it lists, its Storage collection, then each storage resource, each of its Drives, and each member of its Volumes collection.

{"Id": "RAID.Integrated.1-1", "Name": "PERC H755 Front", "Status": {"Health": "OK", "HealthRollup": "Warning"},
 "Drives": [{"@odata.id": "/redfish/v1/Systems/System.Embedded.1/Storage/RAID.Integrated.1-1/Drives/Disk.Bay.0"}],
 "Volumes": {"@odata.id": "/redfish/v1/Systems/System.Embedded.1/Storage/RAID.Integrated.1-1/Volumes"}}

A storage resource is a storage subsystem, most often one controller: it is returned with the kind controller; a drive with drive, a volume with volume.
Each is named by its Name; when two bear the same name (two drives Disk 0 on two controllers), each is named by its name followed by its Id.
health is the resource's own Health, in lower case (ok, warning, critical), unknown when it gives none. Its own, not its HealthRollup: a controller would otherwise raise again for each failed drive under it.
failure-predicted reads yes when the drive says it will fail (FailurePredicted, the drive's SMART verdict as the controller reads it), no otherwise; not returned when the card does not say.
remaining-life is PredictedMediaLifeLeftPercent, for a drive that estimates it (an SSD).
A drive or a volume the card says Absent (an empty bay) is left out.
A code 401 or 403 stops the collector misconfigured; any other code but 200 stops it unreachable, with the path it was reading. A card that serves no storage resource stops it misconfigured.

The password comes from protocol:http:password, a secret variable: it only goes in the Basic header. Redfish only reads with GET.

Prerequisites

The management card's address as the host's address, reachable from the poller in HTTPS (443); its certificate trusted by the TLS profile of the probe (protocol:http:fingerprint, or its authority under protocol:tls:authorities).
A read-only account on the card (the ReadOnly role of Redfish), in protocol:http:user and protocol:http:password.

Outputs

Name	Label	Unit	Per instance
`health`	Health	—	yes, per component
`kind`	Kind	—	yes, per component
`failure-predicted`	Failure predicted	—	yes, per component
`remaining-life`	Remaining life	%	yes, per component

Watches

Name	Title	Kind	Defaults	Checked by default
`unhealthy`	Health degraded	state to avoid, on `health`	warning on `warning`, critical on `critical`	yes
`failure-predicted`	Drive failure predicted	state to avoid, on `failure-predicted`	warning on `yes`	yes
`worn`	Drive worn out	below a threshold, on `remaining-life`	warning below 10 %, critical below 3 %	yes
`missing`	Component missing	expected instances, on `health`	no component named: it raises nothing until names are set on the probe	no

A volume that loses a drive of its mirror reads warning (degraded) while it rebuilds; a volume lost reads critical.

Limits

The address is written into the URL as is: an IPv6 address must be given as a name.
One request per drive and per volume: a server with many drives takes as many round trips.
A controller the card serves only under a resource of its own brand, outside Storage (HPE iLO's SmartStorage for older Smart Array controllers), is not read; a card that serves no storage at all stops the collector misconfigured.
A collection cut into pages (Members@odata.nextLink) is read on its first page only.
A JSON null reads as '' in Keen, which the collector takes for "not given": a PredictedMediaLifeLeftPercent or a FailurePredicted written null is not returned, a Health written null reads unknown, which raises nothing.
Not checked against a card: the paths and fields are those of the DMTF Redfish schemas (ComputerSystem, Storage, Drive, Volume), read on redfish.dmtf.org.