Skip to content

Commit 00c6eb1

Browse files
myasnikovdaniilmyasnikovdaniil
committed
docs(e2e): name the container lane's one-per-host limit
SANDBOX_NAME is per-checkout and these are not, which reads like an oversight until you notice that hack/e2e-compose.yaml's container names and the data-srvN zpools are global too. Two checkouts on one host therefore share nodes and pools, and the second one's teardown destroys the first one's storage. Recorded rather than fixed. CI cannot reach it -- the lane already asks for 24 vCPU and 72 GiB against a 32 vCPU / 128 GiB runner, so a second one does not fit beside it -- and isolating it properly means renaming the pools, which the LINSTOR storage-pool registration reads. Signed-off-by: Myasnikov Daniil <[email protected]> Assisted-By: LLM
1 parent c748c5b commit 00c6eb1

1 file changed

Lines changed: 8 additions & 0 deletions

File tree

‎packages/core/testing/Makefile‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,14 @@ COZY_E2E_NODE_CPUS ?= 8
77
COZY_E2E_NODE_MEMORY_MIB ?= 24576
88

99
ROOT_DIR = $(dir $(abspath $(firstword $(MAKEFILE_LIST))/../../..))
10+
# One container lane per host. Unlike SANDBOX_NAME above, these names are not
11+
# per-checkout, and hack/e2e-compose.yaml's `container_name: srv1..3` and the
12+
# `data-srvN` zpools are global besides -- so a second checkout on the same host
13+
# adopts this one's nodes and pools, and its delete-cluster-container destroys
14+
# them. Not reachable in CI, where the lane already fills the largest runner
15+
# shape (3 x 8 vCPU / 24 GiB against 32 / 128), and left as a known limit rather
16+
# than fixed because isolating it properly means renaming the pools, which the
17+
# LINSTOR storage-pool registration reads.
1018
COMPOSE_PROJECT ?= cozy-e2e
1119
COMPOSE_FILE ?= $(ROOT_DIR)hack/e2e-compose.yaml
1220
ZPOOL_BACKING_DIR ?= /var/lib/cozy-e2e-zpools

0 commit comments

Comments
 (0)