Table of contents
- Quick read
- Readiness check
- List every job that runs without a person
- Identify the owner and the purpose
- Record the credentials and the data each job touches
- Set the failure signal and the retry limit
- Review jobs before a platform change
- Retire a job with evidence
- Publish the register where the team works
- Role brief
- Frequently asked questions
- Sources
Is the handoff ready?
Use these checks before access expands or the first ticket begins.
Job inventory
Ready: List cron, timer, scheduler, and worker jobs from the systems, not memory.
Needs work: The lane has no written record for this check.
Named owner
Ready: Record a primary and backup owner who can explain and change the job.
Needs work: The lane has no written record for this check.
Access record
Ready: Name the identity, scope, secret store, and rotation owner for each job.
Needs work: The lane has no written record for this check.
Failure rule
Ready: Set the alert target, expected duration, retry limit, and safe rerun command.
Needs work: The lane has no written record for this check.
List every job that runs without a person
A scheduled job register starts with the jobs that run without anyone watching them. Include cron entries, systemd timers, cloud schedulers, queue workers, and report generators. Search the servers, the deployment configuration, and the platform consoles rather than asking the team to remember. Record every job that can change data, send a message, or call a paid service.
Note the schedule in plain language and in the system time zone. Record the command or the function the job runs and the repository that contains it. Mark the jobs that were created by a former contractor or a temporary project. Those jobs are the ones most likely to fail without an owner.
Check the container scheduler and the serverless console too, because a job can run without a cron file on any host. Record the platform that owns the schedule. A job that appears in two systems needs one authoritative entry. Mark the entry that the team will update when the schedule changes.
Identify the owner and the purpose
Give every job a named owner who can explain why it exists and what happens if it stops. Record a backup owner who can change the schedule or disable the job. Write one sentence about the business outcome the job supports. A job described only by its command name has no owner in practice.
Separate a job that supports a customer promise from one that only cleans up internal data. Use that distinction when the team has limited time to repair failures. Record the date the purpose was last confirmed. A purpose that nobody can confirm is a candidate for retirement.
Record the credentials and the data each job touches
List the identity each job uses, the permissions it holds, and where the secret is stored. Record the tables, buckets, and external endpoints the job can reach. Mark any job that uses a shared account or a personal token. Those jobs block revocation and hide who ran the work.
Give each job its own account where the platform allows it. Limit the account to the resources the job needs and rotate the secret on a written schedule. Record the rotation owner and the last rotation date. A job token that never rotates is a standing risk.
Set the failure signal and the retry limit
Define what a failed run looks like and where the alert goes. Record the expected duration so a silent hang can be detected. Set a retry limit and the delay between attempts. An unlimited retry can turn one failure into a queue of duplicate work.
Name the person who responds to an alert and the action they take first. Write the safe manual command for rerunning the job. Record how the team prevents a duplicate send or a double charge. The response should be short enough to use at night.
Test the alert by forcing a safe failure on a nonproduction job. Confirm that the message reaches the named responder and contains the job name and environment. Record the test date. An alert that has never fired is a hope.
Review jobs before a platform change
Review the register before an operating system upgrade, a database migration, or a cloud move. Check that every job has a tested path on the new platform. Record the jobs that need a rewrite and the owner of each change. A platform change is the moment hidden jobs fail together.
Test one high-value job in the new environment before the cutover. Compare its output with the current output and record the difference. Keep the old schedule available until the new one has run successfully. Remove the old entry only after the evidence is complete.
Retire a job with evidence
Disable a job by removing its schedule first and watching for unexpected side effects. Record the date, the owner, and the reason for retirement. Keep the code and configuration until the retention period ends. A deleted job that someone still depends on becomes an outage.
Review the register each quarter against the jobs that actually ran. Remove entries that no longer exist and investigate jobs that appear without a record. Publish the updated register to the delivery team. A register that nobody reviews is a list of guesses.
Notify the people who receive the job output before the schedule is removed. Give them a date and a replacement path if one exists. Record their confirmation in the retirement ticket. A silent removal turns a cleanup into a surprise.
Publish the register where the team works
Keep the register in the same workspace the team uses for tickets and runbooks. Link each job to its repository, dashboard, and alert rule. Give read access to support and engineering so nobody needs to ask for the list. A register stored in one person's notes disappears when that person is unavailable.
Publish a short summary each quarter with the jobs added, retired, and still unowned. Show the count and the oldest unowned job. Send it to the client technical owner and the delivery lead. A visible trend is easier to act on than a long table.
Copy-ready brief
Put the operating agreement in writing.
List cron, timer, scheduler, and worker jobs from the systems, not memory.
Record a primary and backup owner who can explain and change the job.
Name the identity, scope, secret store, and rotation owner for each job.
Set the alert target, expected duration, retry limit, and safe rerun command.
Frequently asked questions
Does a team mailbox count as a job owner?
No. Name a person and a backup, because a mailbox cannot make a decision or change a schedule.
How often should the register be reviewed?
Quarterly, and before any platform, database, or cloud migration.
When can a job be deleted?
After its schedule is removed, no dependency appears, and the retention record is complete.
Sources
- crontab manual
Cron schedule fields and job definition behavior.
- systemd.timer manual
Timer units, activation, and scheduling behavior.
- NIST SP 800-53 security and privacy controls
Accountability, least privilege, and configuration control.
- Google SRE Workbook
Operational ownership and toil reduction practice.
