Day 4 - Job Scheduling: From cron's 5 Fields to at and batch

📌 Key Takeaways

• Scheduling tools come in two branches — recurring jobs use cron (crond), one-time jobs use at and batch (atd).
• The crontab 5-field order is "minute (0-59) hour (0-23) day (1-31) month (1-12) weekday (0-7)". For the weekday, both 0 and 7 mean Sunday, and time is on the 24-hour clock.
• The symbols are * (all), , (list), - (range), / (step). If you specify both day and weekday, they work as OR, not AND.
• A user crontab has 5 fields, while a system crontab (/etc/crontab, /etc/cron.d/) has 6 fields with a user field added.
• Access control is allow > deny — if cron.allow exists, only that list is allowed and cron.deny is ignored.

cron's Structure — The crond Daemon and crontab

• cron is a system for scheduling recurring jobs.
• The background daemon crond wakes up every minute, checks every user's schedule list, and runs the jobs whose time has come.
• The file/table holding the schedule list is called the crontab (cron table).
• The command that manages per-user crontabs is crontab.

crontab -e              # add/modify my scheduled jobs in an editor
crontab -l              # view the currently registered scheduled jobs
crontab -r              # delete all scheduled jobs (careful!)
crontab -u alice -l     # view alice's scheduled jobs (root privilege)

⚠️ Trap: crontab -r deletes everything without confirmation. If you mistakenly hit -r, one key away from the -e you meant to press, all your schedules vanish. In practice, the habit of backing up first with crontab -l > backup.txt is recommended. You must clearly distinguish -l (list), -e (edit), and -r (delete).

The crontab's 5 Fields — The Heart of Time Notation

• One crontab line = 5 time fields + the command to run. The order and ranges of these 5 fields are the absolute core of the exam.

┌───────────── minute  (0-59)
│ ┌─────────── hour    (0-23)
│ │ ┌───────── day     (1-31)
│ │ │ ┌─────── month   (1-12)
│ │ │ │ ┌───── weekday (0-7, both 0 and 7 are Sunday)
│ │ │ │ │
* * * * *  command_to_run

• Special characters express patterns.

• Get used to the patterns through real-world examples.

# min hour day month weekday  command
  0  3  *  *  *    /backup.sh        # every day at 3:00 a.m.
  30 2  *  *  1    /weekly.sh        # every Monday at 2:30 a.m.
  0  */6 * *  *    /check.sh         # every 6 hours (at 0, 6, 12, 18)
  */10 * * *  *    /poll.sh          # every 10 minutes
  0  0  1  *  *    /monthly.sh       # midnight on the 1st of every month
  0  9  *  *  1-5  /workday.sh       # 9:00 a.m. on weekdays (Mon-Fri)
  15 14 1  *  *    /report.sh        # 14:15 on the 1st of every month

The order of the 5 fields is "minute-hour-day-month-weekday", from the smaller time unit to the larger (the weekday alone is the exception, placed last). The weekday starts with 0=Sunday and goes to 6=Saturday, and 7 is also accepted as Sunday. Since it is the 24-hour clock, 2 p.m. is 14. Memorize this order with the mnemonic "min-hour-day-month-weekday".

⚠️ Trap: If you specify both the day (3rd) and the weekday (5th), they work as OR — the job runs if either one matches. For example, 0 0 13 * 5 runs "on the 13th of every month or every Friday" (not only on a Friday that is also the 13th), and it is not AND. This OR behavior easily becomes a trap.

Try it yourself: Use crontab -e to put in the single line */2 * * * * date >> ~/crontest.log and save. The current time accumulates in the file every 2 minutes. After a few minutes, check the accumulation with cat ~/crontest.log, and when you're done, delete that line in crontab -e. You get to see how the 5 fields actually run.

System crontab — /etc/crontab and cron.d

• Separately from user crontabs (crontab -e), there are places for system-wide schedules — the /etc/crontab file and the /etc/cron.d/ directory.
• The decisive difference is that there is one more user field.
• The reason: a user crontab runs with the registrant's privileges, but a system crontab must specify which user's privileges to run with.

# /etc/crontab format (6 fields = 5 time fields + user + command)
# min hour day month weekday  user  command
  0  4  *  *  *    root    /usr/local/bin/cleanup.sh

• There are also fixed per-interval directories. You don't write a time; just put a script in and it runs automatically.

The difference that a user crontab has 5 fields and a system crontab (/etc/crontab, cron.d) has 6 fields (user added) is a key exam point. /etc/cron.d/ is a good place for packages to add their own schedules, and it uses the same 6-field format as /etc/crontab. Directories like cron.daily take scripts without a time (the run time is managed by anacron/the system).

The word cron comes from the Greek 'chronos' (time). It has existed since 1970s Unix, and anacron was added for systems that aren't always on, like laptops — cron skips a job if the machine is off at that exact time, but anacron runs it "even right after power-on if N days have passed since the last run". This is why cron.daily and the like are tied to anacron.

Access Control — cron.allow and cron.deny

• Who can use cron is controlled by two files.

• Decision rules:
  • If cron.allow exists, only that list is allowed and cron.deny is ignored.
  • If cron.allow does not exist and only cron.deny does, everyone except the deny list is allowed.
  • If neither exists, usually only root (or everyone, depending on the distribution) is allowed.

The priority is allow > deny. The moment an allow file exists it becomes the only criterion and deny loses effect. at is controlled in the same way by /etc/at.allow and /etc/at.deny.

Run Just Once — at and batch

• If cron is "recurring", at is "exactly once". It runs a command once at the specified time and then disappears from the queue.

at 17:00                  # input mode for a command to run at 5 p.m.
at> /usr/local/bin/report.sh
at> <end input with Ctrl+D>
 
at now + 2 hours          # 2 hours from now
at 10:30 tomorrow         # 10:30 a.m. tomorrow
at midnight               # midnight today
echo "/backup.sh" | at 3am  # pass the command via a pipe

• There are separate commands for managing at jobs.

• batch is a variant of at. It doesn't specify a time and runs automatically when the system load (load average) drops (usually below 0.8).
• Purpose: when you want to run heavy jobs while the CPU is idle.

batch                     # runs when the system becomes idle
at> /heavy-compute.sh
at> <Ctrl+D>

cron is recurring, at/batch are one-time. "Back up every day at dawn" is cron, "one-time inspection tomorrow" is at. at ends its input with Ctrl+D (EOF), the list of registered jobs is atq, and deletion is atrm (at -l/at -d work the same). The difference from at is that batch runs based on load rather than time.

Try it yourself: Run echo "echo hello > ~/at-test.txt" | at now + 1 minute and check the pending list with atq. If cat ~/at-test.txt shows hello a minute later, it worked, and the job disappears from atq (confirming it is one-time). To cancel, look up the number with atq and run atrm number.

📖 Glossary

• crond : The daemon that wakes up every minute, inspects the schedule list, and runs jobs whose time has come.
• crontab : The table holding the schedule list. Handled with -e (edit), -l (list), -r (delete all, no confirmation).
• 5 fields : Time notation in the order minute-hour-day-month-weekday. Memorize it as "min-hour-day-month-weekday".
• step (/) : The interval symbol. */10 means "every 10".
• Day/weekday OR rule : If you fill in both the 3rd (day) and 5th (weekday), the job runs if either one matches.
• System crontab : /etc/crontab and /etc/cron.d/*. An execution user field follows the 5 time fields, making 6 fields.
• /etc/cron.daily/ etc. : Directories where, if you just put a script with no time, it runs at that interval.
• anacron : For systems that aren't always on. It runs jobs cron skipped "even right after power-on".
• cron.allow / cron.deny : Usage-permission whitelist / prohibition blacklist. If allow exists, deny is ignored.
• at : A one-time schedule that runs exactly once at the specified time and then disappears from the queue. List with atq, delete with atrm.
• batch : A variant of at that runs once when the system load drops, not at a given time.
• atd : The daemon that handles at and batch jobs. It is separate from crond.