Skip to content

Commit ee50f94

Browse files
samrosesamrose
committed
test: use the right alias on machine
1 parent e857e9d commit ee50f94

File tree

3 files changed

+14
-144
lines changed

3 files changed

+14
-144
lines changed

‎.github/workflows/testinfra-ami-build.yml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -108,13 +108,13 @@ jobs:
108108
df -h / # Display available space
109109
110110
- name: Run tests
111-
timeout-minutes: 10
111+
timeout-minutes: 30
112112
env:
113113
AMI_NAME: "supabase-postgres-${{ steps.random.outputs.random_string }}"
114114
run: |
115115
# TODO: use poetry for pkg mgmt
116116
pip3 install boto3 boto3-stubs[essential] docker ec2instanceconnectcli pytest pytest-testinfra[paramiko,docker] requests
117-
pytest -vv -s testinfra/test_ami_nix.py
117+
pytest -vvvv -s testinfra/test_ami_nix.py
118118
119119
- name: Cleanup resources on build cancellation
120120
if: ${{ cancelled() }}

‎ansible/files/postgres_prestart.sh.j2

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -41,7 +41,7 @@ get_pg_cron_version() {
4141
fi
4242

4343
local version
44-
version=$(sudo -u postgres /home/postgres/.nix-profile/bin/jq -r '.pg_cron // empty' "/root/pg_extensions.json")
44+
version=$(sudo -u postgres /var/lib/postgresql/.nix-profile/bin/jq -r '.pg_cron // empty' "/root/pg_extensions.json")
4545
if [ -z "$version" ]; then
4646
echo "pg_cron: Not specified in extensions file"
4747
return
@@ -57,7 +57,7 @@ get_pg_cron_version() {
5757

5858
switch_pg_cron_version() {
5959
local version="$1"
60-
local switch_script="/home/postgres/.nix-profile/bin/switch_pg_cron_version"
60+
local switch_script="/var/lib/postgresql/.nix-profile/bin/switch_pg_cron_version"
6161

6262
if [ ! -x "$switch_script" ]; then
6363
echo "pg_cron: No version switch script available"

‎testinfra/test_ami_nix.py

Lines changed: 10 additions & 140 deletions
Original file line numberDiff line numberDiff line change
@@ -348,171 +348,41 @@ def gzip_then_base64_encode(s: str) -> str:
348348

349349
def is_healthy(ssh) -> bool:
350350
health_checks = [
351-
<<<<<<< HEAD
352351
("postgres", "sudo -u postgres /usr/bin/pg_isready -U postgres"),
353352
("adminapi", f"curl -sf -k --connect-timeout 30 --max-time 60 https://localhost:8085/health -H 'apikey: {supabase_admin_key}'"),
354353
("postgrest", "curl -sf --connect-timeout 30 --max-time 60 http://localhost:3001/ready"),
355354
("gotrue", "curl -sf --connect-timeout 30 --max-time 60 http://localhost:8081/health"),
356355
("kong", "sudo kong health"),
357356
("fail2ban", "sudo fail2ban-client status"),
358-
=======
359-
(
360-
"postgres",
361-
lambda h: (
362-
# First check if PostgreSQL is running
363-
h.run("sudo systemctl is-active postgresql"),
364-
# Then check if the socket directory exists and has correct permissions
365-
h.run("sudo ls -la /run/postgresql"),
366-
# Then try pg_isready
367-
h.run("sudo -u postgres /usr/bin/pg_isready -U postgres")
368-
),
369-
),
370-
(
371-
"adminapi",
372-
lambda h: h.run(
373-
f"curl -sf -k --connect-timeout 30 --max-time 60 https://localhost:8085/health -H 'apikey: {supabase_admin_key}'"
374-
),
375-
),
376-
(
377-
"postgrest",
378-
lambda h: h.run(
379-
"curl -sf --connect-timeout 30 --max-time 60 http://localhost:3001/ready"
380-
),
381-
),
382-
(
383-
"gotrue",
384-
lambda h: h.run(
385-
"curl -sf --connect-timeout 30 --max-time 60 http://localhost:8081/health"
386-
),
387-
),
388-
("kong", lambda h: h.run("sudo kong health")),
389-
("fail2ban", lambda h: h.run("sudo fail2ban-client status")),
390-
>>>>>>> 2bd7b6d9 (test: more logging for healthcheck)
391357
]
392358

393359
for service, command in health_checks:
394360
try:
395-
<<<<<<< HEAD
396361
result = run_ssh_command(ssh, command)
397362
if not result['succeeded']:
398363
logger.warning(f"{service} not ready")
399364
logger.error(f"{service} command failed with rc={cmd.rc}")
400365
logger.error(f"{service} stdout: {cmd.stdout}")
401366
logger.error(f"{service} stderr: {cmd.stderr}")
402-
=======
403-
if service == "postgres":
404-
# For PostgreSQL, we need to check multiple things
405-
systemd_status, socket_check, pg_isready = check(host)
406-
>>>>>>> 2bd7b6d9 (test: more logging for healthcheck)
407367

408-
# Log Nix profile setup checks
409-
logger.info("Checking Nix profile setup:")
410-
nix_profile_result = host.run("ls -la /home/postgres/.nix-profile")
411-
logger.info(f"Nix profile directory:\n{nix_profile_result.stdout}\n{nix_profile_result.stderr}")
412-
413-
nix_bin_result = host.run("ls -la /home/postgres/.nix-profile/bin")
414-
logger.info(f"Nix profile bin directory:\n{nix_bin_result.stdout}\n{nix_bin_result.stderr}")
415-
416-
nix_script_result = host.run("test -x /home/postgres/.nix-profile/bin/switch_pg_cron_version")
417-
logger.info(f"Switch script executable check: {'success' if not nix_script_result.failed else 'failed'}")
418-
419-
nix_script_output = host.run("/home/postgres/.nix-profile/bin/switch_pg_cron_version")
420-
logger.info(f"Switch script output:\n{nix_script_output.stdout}\n{nix_script_output.stderr}")
421-
422368
if systemd_status.failed:
423369
logger.error("PostgreSQL systemd service is not active")
424370
logger.error(f"systemd status: {systemd_status.stdout}")
425371
logger.error(f"systemd error: {systemd_status.stderr}")
426-
427-
# Check systemd service unit file
428-
logger.error("PostgreSQL systemd service unit file:")
429-
result = host.run("sudo systemctl cat postgresql")
430-
logger.error(f"service unit file:\n{result.stdout}\n{result.stderr}")
431-
432-
# Check systemd service environment
433-
logger.error("PostgreSQL systemd service environment:")
434-
result = host.run("sudo systemctl show postgresql")
435-
logger.error(f"service environment:\n{result.stdout}\n{result.stderr}")
436-
437-
# Check systemd service dependencies
438-
logger.error("PostgreSQL systemd service dependencies:")
439-
result = host.run("sudo systemctl list-dependencies postgresql")
440-
logger.error(f"service dependencies:\n{result.stdout}\n{result.stderr}")
441-
442-
# Check if service is enabled
443-
logger.error("PostgreSQL service enabled status:")
444-
result = host.run("sudo systemctl is-enabled postgresql")
445-
logger.error(f"service enabled status:\n{result.stdout}\n{result.stderr}")
446-
447-
# Check systemd journal for service execution logs
448-
logger.error("Systemd journal entries for PostgreSQL service execution:")
449-
result = host.run("sudo journalctl -u postgresql -n 100 --no-pager")
450-
logger.error(f"systemd journal:\n{result.stdout}\n{result.stderr}")
451-
452-
# Check systemd journal specifically for ExecStartPre and ExecStart
453-
logger.error("Systemd journal entries for ExecStartPre and ExecStart:")
454-
result = host.run("sudo journalctl -u postgresql -n 100 --no-pager | grep -E 'ExecStartPre|ExecStart'")
455-
logger.error(f"execution logs:\n{result.stdout}\n{result.stderr}")
456-
457-
# Check systemd journal for any errors
458-
logger.error("Systemd journal entries with error level:")
459-
result = host.run("sudo journalctl -u postgresql -n 100 --no-pager -p err")
460-
logger.error(f"error logs:\n{result.stdout}\n{result.stderr}")
461-
462-
# Check pre-start script output
463-
logger.error("Checking pre-start script output:")
464-
result = host.run("sudo -u postgres /usr/local/bin/postgres_prestart.sh")
465-
logger.error(f"pre-start script output:\n{result.stdout}\n{result.stderr}")
466372

467-
# Check PostgreSQL logs directory
468-
logger.error("Checking PostgreSQL logs directory:")
469-
result = host.run("sudo ls -la /var/log/postgresql/")
470-
logger.error(f"log directory contents:\n{result.stdout}\n{result.stderr}")
471-
472-
# Check any existing PostgreSQL logs
473-
logger.error("Checking existing PostgreSQL logs:")
474-
result = host.run("sudo cat /var/log/postgresql/*.log")
475-
logger.error(f"postgresql logs:\n{result.stdout}\n{result.stderr}")
476-
477-
# Try starting PostgreSQL directly with pg_ctl and capture output
478-
logger.error("Attempting to start PostgreSQL directly with pg_ctl:")
479-
startup_log = "/tmp/postgres-start.log"
480-
result = host.run(f"sudo -u postgres /usr/lib/postgresql/bin/pg_ctl -D /var/lib/postgresql/data start -l {startup_log}")
481-
logger.error(f"pg_ctl start attempt:\n{result.stdout}\n{result.stderr}")
482-
483-
# Check the startup log
484-
logger.error("PostgreSQL startup log:")
485-
result = host.run(f"sudo cat {startup_log}")
486-
logger.error(f"startup log contents:\n{result.stdout}\n{result.stderr}")
487-
488-
# Clean up the startup log
489-
result = host.run(f"sudo rm -f {startup_log}")
490-
491-
# Check PostgreSQL configuration
492-
logger.error("PostgreSQL configuration:")
493-
result = host.run("sudo cat /etc/postgresql/postgresql.conf")
494-
logger.error(f"postgresql.conf:\n{result.stdout}\n{result.stderr}")
495-
496-
# Check PostgreSQL authentication configuration
497-
logger.error("PostgreSQL authentication configuration:")
498-
result = host.run("sudo cat /etc/postgresql/pg_hba.conf")
499-
logger.error(f"pg_hba.conf:\n{result.stdout}\n{result.stderr}")
500-
501-
# Check PostgreSQL environment
502-
logger.error("PostgreSQL environment:")
503-
result = host.run("sudo -u postgres env | grep POSTGRES")
504-
logger.error(f"postgres environment:\n{result.stdout}\n{result.stderr}")
373+
# Run detailed checks since we know we have a working connection
374+
run_detailed_checks(host)
505375

506376
if any(cmd.failed for cmd in [systemd_status, socket_check, pg_isready]):
507377
return False
508-
else:
509-
cmd = check(host)
510-
if cmd.failed is True:
511-
logger.warning(f"{service} not ready")
512-
logger.error(f"{service} command failed with rc={cmd.rc}")
513-
logger.error(f"{service} stdout: {cmd.stdout}")
514-
logger.error(f"{service} stderr: {cmd.stderr}")
515-
return False
378+
else:
379+
cmd = check(host)
380+
if cmd.failed is True:
381+
logger.warning(f"{service} not ready")
382+
logger.error(f"{service} command failed with rc={cmd.rc}")
383+
logger.error(f"{service} stdout: {cmd.stdout}")
384+
logger.error(f"{service} stderr: {cmd.stderr}")
385+
return False
516386
except Exception as e:
517387
logger.warning(
518388
f"Connection failed during {service} check, attempting reconnect..."

0 commit comments

Comments
 (0)