Skip to content

Commit 498fac8

Browse files
test(e2e): size the container pool on its own terms, not the QEMU lane's
The container lane's ZPOOL_SIZE comment and the guard beside it both justified 200 GB as a match for the QEMU lane's per-node data disk. This branch moves that disk to 300 GB, because the golden-image clone path holds the clone source and the growing target at once and a `replicated` StorageClass stores both per replica. The parity claim is now false in the one direction that matters: a reader following it would size this lane up for a cost it does not pay. This lane runs drbd.enabled=false with COZY_E2E_STORAGE_CLASS=local, so a worker disk is stored once and the same suites peak several times lower here. 200 GB stays, argued from that rather than from a figure that moved, and the guard's name and message say so too. Assisted-By: LLM Signed-off-by: Myasnikov Daniil <[email protected]>
1 parent 8fbe776 commit 498fac8

2 files changed

Lines changed: 12 additions & 5 deletions

File tree

‎hack/container-lane-capacity_test.bats‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -28,9 +28,9 @@ E2E_CONTAINER_UP_LIB=true . "$CONTAINER_UP"
2828
fi
2929
}
3030

31-
@test "container storage pools match the QEMU lane capacity" {
31+
@test "container storage pools keep the per-node 200G this lane sizes against" {
3232
if ! grep -Fq 'ZPOOL_SIZE="${ZPOOL_SIZE:-200G}"' "$CONTAINER_UP"; then
33-
echo "container ZFS pools do not provide the QEMU lane's 200G per node" >&2
33+
echo "container ZFS pools do not provide the 200G per node this lane sizes against" >&2
3434
return 1
3535
fi
3636
}

‎hack/e2e-container-up.sh‎

Lines changed: 10 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -111,9 +111,16 @@ COMPOSE_PROJECT="${COMPOSE_PROJECT:-cozy-e2e}"
111111
# A smaller pool is not safe even when its sparse backing file fits physically:
112112
# root-tenant local PVCs can leave a satellite with less than the ~21 GiB needed
113113
# by CDI scratch. Once a worker's 20 GiB target PV is bound there, node affinity
114-
# forces the importer and scratch volume onto that same full satellite. Match the
115-
# QEMU lane's per-node 200 GB /dev/vdc so placement cannot turn node join into a
116-
# storage-capacity lottery.
114+
# forces the importer and scratch volume onto that same full satellite, and
115+
# placement turns node join into a storage-capacity lottery.
116+
#
117+
# 200 GB is this lane's own figure and no longer the QEMU lane's, which is larger.
118+
# DRBD is what separates them: this lane runs drbd.enabled=false with
119+
# COZY_E2E_STORAGE_CLASS=local, so a worker disk is stored once, while the QEMU
120+
# lane's `replicated` class stores it per replica. The golden-image clone path
121+
# holds the clone source and the growing target at the same time, so the same
122+
# suites peak several times higher over there -- see the sizing note in
123+
# hack/e2e-prepare-cluster.bats. Size this lane against a single copy.
117124
ZPOOL_SIZE="${ZPOOL_SIZE:-200G}"
118125
ZPOOL_BACKING_DIR="${ZPOOL_BACKING_DIR:-/var/lib/cozy-e2e-zpools}"
119126
KUBERNETES_VERSION="${KUBERNETES_VERSION:-v1.33.12}"

0 commit comments

Comments
 (0)