Files
cozystack/packages/system/linstor/values.yaml
T

122 lines
6.7 KiB
YAML

piraeusServer:
image:
repository: ghcr.io/cozystack/cozystack/piraeus-server
tag: v1.6.0@sha256:70ddd1a66fbc1ea58078f4645fc572ab8e2e189dbb4d9a5f9ca43d5663106483
# Talos-specific workarounds (disable for generic Linux like Ubuntu/Debian)
talos:
enabled: true
gracefulShutdown:
# Release unused Secondary DRBD resources before satellite termination so
# ZFS can release their backing zvols during normal Talos shutdown.
# The hook also runs on satellite pod restarts. Enable only when satellite
# and node restarts are serialized: Piraeus has one DaemonSet per node, so
# a shared configuration change can otherwise stop several replicas at once.
enabled: false
# Whether the node substrate can run DRBD at all.
#
# Disable on any substrate where the DRBD kernel module is unavailable or
# unusable. The concrete case is Talos nodes running as containers: DRBD's
# resource registry is global per kernel, so three containers sharing one kernel
# cannot represent three DRBD nodes — the second `new-resource` for a given name
# silently aliases the first instead of failing, which is worse than an error
# because LINSTOR records a replica it does not have.
#
# What this flag does, and all it does: the `drbd-logger` sidecar is omitted from
# Cozystack's sparse pod template and deleted explicitly from the effective
# DaemonSet. Omission alone is not enough because piraeus strategically merges
# satellite configurations. That sidecar tails DRBD state and exits immediately
# when the module is absent, and because the whole satellite pod is then never
# Ready, piraeus never registers the node: LINSTOR reports zero nodes,
# `linstor-csi-node` hangs in Init polling for a node that does not exist, and
# every PVC stays Pending. So this flag is the difference between working storage
# and no storage at all on such a substrate, not a cosmetic reduction in logging.
#
# What it does NOT do: it does not touch a StorageClass. Cozystack does not own
# them — an operator writes them by hand from the install guide — so a class
# carrying `linstor.csi.linbit.com/layerList: "drbd storage"` still exists after
# setting this false, and volumes on it will never provision. On such a substrate
# create only storage-only, node-pinned classes (`layerList: "storage"`,
# `allowRemoteVolumeAccess: "false"`); the install docs say so beside the
# examples.
#
# What it REQUIRES: talos.enabled must stay true. Removing the sidecar is not the
# whole job — the satellite DaemonSet also carries two DRBD-only initContainers
# that piraeus-operator contributes unconditionally (drbd-module-loader, which
# fails outright when the module is absent, and drbd-shutdown-guard), and the
# only configuration that deletes them is the Talos one. On generic Linux the
# satellite would therefore still never reach Ready, so the chart refuses the
# combination at render time rather than shipping a DaemonSet that cannot come
# up. A DRBD-less substrate has to be Talos today; making generic Linux work
# means moving those two deletions out from under talos.enabled, tracked in
# cozystack/cozystack#4092.
#
# Leaving it true on a DRBD-capable node is always correct; this only ever needs
# disabling where DRBD genuinely cannot work.
drbd:
enabled: true
linstor:
autoDiskful:
enabled: true
minutes: 30
allowCleanup: true
linstorCSI:
## Constrain provisioning to the node the scheduler actually picked.
##
## Without it, external-provisioner passes EVERY topology segment as
## `requisite` and merely *prefers* the selected node, so when that node is out
## of space linstor-csi may legally provision the volume somewhere else. For a
## node-pinned volume (`allowRemoteVolumeAccess: false`, the `local` class that
## is the cluster default) the result is a PV whose node affinity contradicts
## the pod's other volumes, and the pod is then unschedulable FOREVER:
##
## pv "pvc-..." node affinity doesn't match node "srv1": no matching NodeSelectorTerms
##
## It bites hardest on CDI imports, because an importer pod mounts two volumes
## -- the disk and a ~21 GiB scratch volume that CDI only creates AFTER the pod
## is scheduled. The scheduler therefore sizes the node against the disk alone,
## can place two importers on a node that cannot hold both disks plus both
## scratch volumes, and the second scratch is then placed elsewhere.
##
## Measured on the wire, on a three-node dev stand: the CreateVolume request
## for a WaitForFirstConsumer volume carries `requisite` as the single segment
## the scheduler chose when the flag is set, and all three node segments when
## it is not, the chosen one merely first in `preferred`.
##
## Reproduced and fixed by measurement 2026-08-25 on a three-node cluster: with
## the flag absent the scratch PV landed on a different node than its disk and
## the import hung indefinitely at ImportScheduled with no error anywhere; with
## it set the scratch PVC stays Pending and the pod reports the actionable
## `1 node(s) did not have enough free storage`, then binds and completes on
## its own once space frees up. A silent permanent deadlock becomes a legible,
## self-correcting wait.
##
## Verified for node-pinned (`local`) volumes, which is what the cluster
## default uses. `replicated` is untouched for a sharper reason than its PV
## being remotely accessible: it binds `Immediate`, and external-provisioner
## scopes strict topology to late binding, so the flag never reaches that
## path at all. The switch stays because the flag is an upstream default this
## chart overrides, not because the other class is in doubt.
strictTopology: true
image:
repository: ghcr.io/cozystack/cozystack/linstor-csi
tag: v1.6.0@sha256:ae854423de62e94fb89b59d92c20da8939196e643e83b2d60d9f6915d5b0e55c
controller:
# Resource requests for the linstor-controller JVM. Requests (not limits) are
# set on purpose: a CPU request guarantees CPU shares so the JVM can answer its
# GET /health liveness probe under node CPU contention, while omitting limits
# avoids throttling the JVM (CPU limit) or OOM-killing it (memory limit). The
# memory value is a conservative starting estimate for the JVM heap + overhead;
# tune it to observed RSS.
resources:
requests:
cpu: 250m
memory: 700Mi
# Liveness-probe timing for the linstor-controller container. Only these timing
# fields are overridden; the httpGet target (GET /health on the api port) is
# supplied by piraeus-operator and intentionally left untouched. Relaxed from
# the operator defaults (timeoutSeconds 1, failureThreshold 3) so a JVM GC
# pause or CPU contention does not trip a kubelet liveness kill.
livenessProbe:
timeoutSeconds: 5
failureThreshold: 6