ci/lava: Use CI_JOB_TIMEOUT instead of separate variable

The CI_JOB_TIMEOUT variable is the GitLab-defined job timeout in
seconds.
Use this variable in LAVA instead of the separate JOB_TIMEOUT,
which was intended to represent the test phase timeout (job timeout
minus 5 minutes), but was often overlooked.

Signed-off-by: Valentine Burley <valentine.burley@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32609>

.gitlab-ci/lava/lava-submit.sh

@@ -69,7 +69,7 @@ PYTHONPATH=artifacts/ artifacts/lava/lava_job_submitter.py \
 	--farm "${FARM}" \
 	--device-type "${DEVICE_TYPE}" \
 	--boot-method "${BOOT_METHOD}" \
-	--job-timeout-min ${JOB_TIMEOUT:-30} \
+	--job-timeout-min $((CI_JOB_TIMEOUT/60 - 5)) \
 	--dump-yaml \
 	--pipeline-info "$CI_JOB_NAME: $CI_PIPELINE_URL on $CI_COMMIT_REF_NAME ${CI_NODE_INDEX}/${CI_NODE_TOTAL}" \
 	--rootfs-url "${ROOTFS_URL}" \

.gitlab-ci/lava/utils/log_section.py

@@ -35,9 +35,13 @@ LAVA_QUEUE_TIMEOUT = int(getenv("LAVA_QUEUE_TIMEOUT", 60))
 # the enqueue delay.
 LAVA_BOOT_TIMEOUT = int(getenv("LAVA_BOOT_TIMEOUT", 9))
 
+# Estimated overhead in minutes for a job from GitLab to reach the test phase,
+# including LAVA scheduling and boot duration
+LAVA_TEST_OVERHEAD_MIN = 5
+
 # Test DUT suite phase is where the initialization happens in DUT, not on docker.
 # The device will be listening to SSH session until the end of the job.
-LAVA_TEST_DUT_SUITE_TIMEOUT = int(getenv("JOB_TIMEOUT", 60))
+LAVA_TEST_DUT_SUITE_TIMEOUT = int(getenv("CI_JOB_TIMEOUT")) // 60 - LAVA_TEST_OVERHEAD_MIN
 
 # Test suite phase is where the initialization happens on docker.
 LAVA_TEST_SUITE_TIMEOUT = int(getenv("LAVA_TEST_SUITE_TIMEOUT", 5))
@@ -45,7 +49,7 @@ LAVA_TEST_SUITE_TIMEOUT = int(getenv("LAVA_TEST_SUITE_TIMEOUT", 5))
 # Test cases may take a long time, this script has no right to interrupt
 # them. But if the test case takes almost 1h, it will never succeed due to
 # Gitlab job timeout.
-LAVA_TEST_CASE_TIMEOUT = int(getenv("JOB_TIMEOUT", 60))
+LAVA_TEST_CASE_TIMEOUT = int(getenv("CI_JOB_TIMEOUT")) // 60 - LAVA_TEST_OVERHEAD_MIN
 
 # LAVA post processing may refer to a test suite teardown, or the
 # adjustments to start the next test_case

src/amd/ci/gitlab-ci.yml

@@ -29,7 +29,6 @@ radv-stoney-angle-full:
   parallel: 2
   variables:
     DEQP_SUITE: radv-stoney-angle-full
-    JOB_TIMEOUT: 60
 
 radeonsi-stoney-gl:
   extends:
@@ -149,7 +148,6 @@ radeonsi-raven-va-full:
     - radeonsi-raven-va
     - .radeonsi-vaapi-manual-rules
   variables:
-    JOB_TIMEOUT: 60
     GTEST_FRACTION: null
 
 ############### VKCTS tests ###############

src/freedreno/ci/gitlab-ci.yml

@@ -103,7 +103,6 @@ a618-vk-full:
     # ran into OOM with VK-GL-CTS 1.2.8.0 at 6
     FDO_CI_CONCURRENT: 4
     DEQP_SUITE: freedreno-a618-vk-full
-    JOB_TIMEOUT: 115
     DEQP_FRACTION: 1
 
 a618-gl:
@@ -123,7 +122,6 @@ a618-gl-full:
   timeout: 65m
   variables:
     DEQP_FRACTION: 1
-    JOB_TIMEOUT: 60
 
 # Run dEQP EGL window system tests separately with the window systems available.
 # X11 takes over the screen, wayland is run headless.
@@ -173,7 +171,6 @@ a618-piglit-full:
     DEQP_SUITE: freedreno-a618-piglit-full
     FLAKES_CHANNEL: "#freedreno-ci"
     HWCI_START_WESTON: 1
-    JOB_TIMEOUT: 60
 
 a618-piglit-cl:
   extends:
@@ -261,7 +258,6 @@ a660-vk-full:
   timeout: 1h
   variables:
     DEQP_SUITE: freedreno-a660-vk-full
-    JOB_TIMEOUT: 55
 
 # X11 takes over the screen, wayland is run headless.
 a630-gl:

src/gallium/drivers/zink/ci/gitlab-ci.yml

@@ -57,7 +57,6 @@ zink-anv-adl-full:
     - .zink-anv-manual-rules
   variables:
     DEQP_SUITE: zink-anv-adl-full
-    JOB_TIMEOUT: 105
     HWCI_START_WESTON: 1
   timeout: 2h
   parallel: 2
@@ -82,7 +81,6 @@ zink-anv-tgl-full:
     - .zink-anv-manual-rules
   variables:
     DEQP_SUITE: zink-anv-tgl-full
-    JOB_TIMEOUT: 105
     HWCI_START_WESTON: 1
   timeout: 1h 45m
   parallel: 3

src/intel/ci/gitlab-ci.yml

@@ -16,7 +16,6 @@ anv-jsl-full:
     - .anv-manual-rules
   variables:
     DEQP_SUITE: anv-jsl-full
-    JOB_TIMEOUT: 115
   parallel: 4
   timeout: 2h
 
@@ -36,7 +35,6 @@ anv-adl-full:
     - .anv-manual-rules
   variables:
     DEQP_SUITE: anv-adl-full
-    JOB_TIMEOUT: 90
   timeout: 1h 40m
   parallel: 5
 
@@ -55,7 +53,6 @@ anv-tgl-full:
     - .anv-manual-rules
   variables:
     DEQP_SUITE: anv-tgl-full
-    JOB_TIMEOUT: 90
   timeout: 100m
   parallel: 4
 
@@ -94,7 +91,6 @@ anv-jsl-angle-full:
     - .intel-manual-rules
   variables:
     DEQP_SUITE: anv-jsl-angle-full
-    JOB_TIMEOUT: 60
   parallel: 2
 
 anv-adl-angle:
@@ -113,7 +109,6 @@ anv-adl-angle-full:
     - .intel-manual-rules
   variables:
     DEQP_SUITE: anv-adl-angle-full
-    JOB_TIMEOUT: 60
   timeout: 30m
 
 anv-tgl-angle:
@@ -131,7 +126,6 @@ anv-tgl-angle-full:
     - .intel-manual-rules
   variables:
     DEQP_SUITE: anv-tgl-angle-full
-    JOB_TIMEOUT: 60
 
 iris-apl-deqp:
   extends:

src/panfrost/ci/gitlab-ci.yml

@@ -262,7 +262,6 @@ panfrost-t760-gles:arm32:
     - .panfrost-test
     - .lava-rk3288-veyron-jaq:arm32
   variables:
-    JOB_TIMEOUT: 60
 
 panfrost-t760-traces:arm32:
   extends:
@@ -354,7 +353,6 @@ panfrost-g52-vk-full:arm64:
   timeout: 3h
   variables:
     DEQP_SUITE: panfrost-g52-vk-full
-    JOB_TIMEOUT: 185
 
 panfrost-g52-piglit-gles2:arm64:
   extends:
@@ -430,4 +428,3 @@ panfrost-g610-vk-full:arm64:
   timeout: 1h 35m
   variables:
     DEQP_SUITE: panfrost-g610-vk-full
-    JOB_TIMEOUT: 90