disinto-admin/disinto
1
0
Fork 0
Code Issues 17 Pull requests 1 Projects Releases Packages Wiki Activity Actions

fix: [nomad-step-1] deploy.sh-fix — poll deployment status not alloc status; bump timeout 120→240s (#878) #885

Merged
dev-qwen merged 2 commits from fix/issue-878 into main 2026-04-16 15:54:59 +00:00
Conversation 4 Commits 2 Files changed 1 +18 -14
1 changed files with 18 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files
Showing only changes of commit 3734920c0c - Show all commits

28
lib/init/nomad/deploy.sh
View file

@ -78,12 +78,21 @@ _wait_job_running() {
log "waiting for job '${job_name}' to become healthy (timeout: ${timeout}s)..." log "waiting for job '${job_name}' to become healthy (timeout: ${timeout}s)..."
# Get the latest deployment ID for this job # Get the latest deployment ID for this job (retry until available)
local deployment_id local deployment_id=""
local retry_count=0
local max_retries=12
while [ -z "$deployment_id" ] && [ "$retry_count" -lt "$max_retries" ]; do
deployment_id=$(nomad job deployments -json "$job_name" 2>/dev/null | jq -r '.[0].ID' 2>/dev/null) || deployment_id="" deployment_id=$(nomad job deployments -json "$job_name" 2>/dev/null | jq -r '.[0].ID' 2>/dev/null) || deployment_id=""
if [ -z "$deployment_id" ]; then
sleep 5
retry_count=$((retry_count + 1))
fi
done
if [ -z "$deployment_id" ]; then if [ -z "$deployment_id" ]; then
log "ERROR: no deployment found for job '${job_name}'" log "ERROR: no deployment found for job '${job_name}' after ${max_retries} attempts"
return 1 return 1
fi fi
@ -99,7 +108,7 @@ _wait_job_running() {
} }
local status local status
status=$(printf '%s' "$deploy_status_json" | jq -r '.[0].Status' 2>/dev/null) || { status=$(printf '%s' "$deploy_status_json" | jq -r '.Status' 2>/dev/null) || {
sleep 5 sleep 5
elapsed=$((elapsed + 5)) elapsed=$((elapsed + 5))
continue continue
@ -114,15 +123,10 @@ _wait_job_running() {
log "deployment '${deployment_id}' failed for job '${job_name}'" log "deployment '${deployment_id}' failed for job '${job_name}'"
log "showing last 50 lines of allocation logs (stderr):" log "showing last 50 lines of allocation logs (stderr):"
# Get allocation IDs from the deployment # Get allocation IDs from job status
local alloc_ids local alloc_ids
alloc_ids=$(printf '%s' "$deploy_status_json" | jq -r '.[0].AllocStatus.AllocsNotYetRunning // empty' 2>/dev/null) || alloc_ids=""
# Fallback: get allocs from job status
if [ -z "$alloc_ids" ]; then
alloc_ids=$(nomad job status -json "$job_name" 2>/dev/null \ alloc_ids=$(nomad job status -json "$job_name" 2>/dev/null \
| jq -r '.Evaluations[].Allocations[]?.ID // empty' 2>/dev/null) || alloc_ids="" | jq -r '.Allocations[]?.ID // empty' 2>/dev/null) || alloc_ids=""
fi
if [ -n "$alloc_ids" ]; then if [ -n "$alloc_ids" ]; then
for alloc_id in $alloc_ids; do for alloc_id in $alloc_ids; do
@ -152,7 +156,7 @@ _wait_job_running() {
# Get allocation IDs from job status # Get allocation IDs from job status
local alloc_ids local alloc_ids
alloc_ids=$(nomad job status -json "$job_name" 2>/dev/null \ alloc_ids=$(nomad job status -json "$job_name" 2>/dev/null \
| jq -r '.Evaluations[].Allocations[]?.ID // empty' 2>/dev/null) || alloc_ids="" | jq -r '.Allocations[]?.ID // empty' 2>/dev/null) || alloc_ids=""
if [ -n "$alloc_ids" ]; then if [ -n "$alloc_ids" ]; then
for alloc_id in $alloc_ids; do for alloc_id in $alloc_ids; do