Start with Field Count: 5 vs 6/7
The field count tells you which ecosystem the expression belongs to, and the two syntaxes are not interchangeable:
Standard 5-Field Cron
* * * * *
│ │ │ │ │
│ │ │ │ └── Day of week (0-7, both 0 and 7 = Sunday)
│ │ │ └──── Month (1-12)
│ │ └────── Day of month (1-31)
│ └──────── Hour (0-23)
└─────────── Minute (0-59)
Quartz 6/7-Field Cron
* * * * * * *
│ │ │ │ │ │ │
│ │ │ │ │ │ └── Year (optional, 1970-2099)
│ │ │ │ │ └──── Day of week (1-7, 1=Sunday)
│ │ │ │ └────── Month (1-12)
│ │ │ └──────── Day of month (1-31)
│ │ └────────── Hour (0-23)
│ └──────────── Minute (0-59)
└─────────────── Second (0-59)
| Field | Standard range | Quartz range | Allowed special chars |
|---|---|---|---|
| Second | — | 0-59 | , - * / |
| Minute | 0-59 | 0-59 | , - * / |
| Hour | 0-23 | 0-23 | , - * / |
| Day | 1-31 | 1-31 | , - * ? L W C |
| Month | 1-12 or JAN-DEC | 1-12 or JAN-DEC | , - * / |
| Day of week | 0-7 or SUN-SAT | 1-7 or SUN-SAT | , - * ? L C # |
| Year | — | 1970-2099 | , - * / |
The difference is not merely two extra fields. In standard cron 0 and 7 are both Sunday; in Quartz 1 is Sunday and 7 is Saturday. Standard cron has no seconds, so the extra leading field is treated as the command: paste 0 0 9 * * ? into a crontab and it fires every day at 09:00 trying to run a command literally named ?. The only trace is command not found — the job looks like it never ran, while it actually fired at the wrong time with the wrong command.
Special Characters in One Table
| Symbol | Meaning | Example | Quartz only? |
|---|---|---|---|
* |
Every legal value in the field | * * * * * every minute |
No |
, |
List of discrete values | 0 0 9,12,18 * * ? 9/12/18 daily |
No |
- |
Inclusive range | 0 0 9-17 * * ? hourly 9:00–17:00 |
No |
/ |
Step, start/step |
*/15 every 15 min; 0/30 at 0 and 30 |
No |
? |
Unspecified, defer to the other day field | 0 0 12 ? * MON Mondays at noon |
Yes |
L |
Last | 0 0 23 L * ? month end; 6L last Friday |
Yes |
W |
Nearest weekday | 15W nearest weekday to the 15th |
Yes |
# |
Nth weekday | 2#1 first Monday; 6#3 third Friday |
Yes |
The Four Easiest Things to Get Wrong
? Is Not "Any"
* means every legal value counts; ? means "I do not specify this field". In Quartz, day-of-month and day-of-week cannot both be concrete, so one must be ?:
0 0 12 ? * MON // every Monday at noon: day unspecified
0 0 12 1 * ? // the 1st at noon: weekday unspecified
Writing ? into a standard 5-field cron is rejected. Conversely, putting * in both day fields in Quartz creates a conflict the scheduler cannot resolve.
L, W, LW: Last Day and Nearest Weekday
L means "last day of the month" in the day field, and "last such weekday" in the weekday field:
0 0 23 L * ? // last day of month at 23:00
0 0 10 ? * 6L // last Friday of month at 10:00
W is day-field only: the nearest weekday to a date. 15W lands on the 14th (Friday) when the 15th is Saturday, and on the 16th (Monday) when it is Sunday. LW means the last weekday of the month.
Standard cron has none of these. The portable "last day" trick is 0 0 28-31 * * plus a script check for whether tomorrow is the 1st. @monthly is the 1st, not the last day — do not confuse them.
Day + Weekday: OR in Standard Cron, ? in Quartz
When both day-of-month and day-of-week are concrete, standard cron combines them with OR: 0 0 1 * 1 fires on the 1st or every Monday, not only when both hold. The symptom is a job that runs more often than intended. If you truly need "the 1st only when it is a Monday", check inside the script.
Quartz forbids the ambiguity: one of the two fields must be ?, so the meaning is always deterministic.
DST Jumps and February 29
In DST regions, the transition day contains an hour that does not exist or one that repeats:
- Spring forward: local 02:00 jumps to 03:00, so
02:00–02:59never happens and a job scheduled there is silently skipped. - Fall back:
02:00–02:59occurs twice, so the same expression may run twice and break logs, de-duplication, and aggregation.
Avoid scheduling critical work between 2:00 and 3:00, or schedule in UTC. Containers default to UTC, so confirm which clock you actually mean.
0 0 0 29 2 * fires only in leap years; for annual work, decide the date in code rather than relying on it.
Common Schedules
| Expression | Meaning |
|---|---|
* * * * * |
Every minute |
*/5 * * * * |
Every 5 minutes |
0 * * * * |
Every hour on the hour |
0 */2 * * * |
Every 2 hours |
0 0 * * * |
Every day at midnight |
0 9 * * * |
Every day at 9 AM |
0 9 * * 1-5 |
Weekdays at 9 AM |
0 22 * * 1-5 |
Weekdays at 10 PM |
0 0 1 * * |
The 1st of each month |
0 0 1 1 * |
January 1st each year |
0 0 * * 0 |
Every Sunday |
0 0 1 */3 * |
First day of each quarter |
0 0 12 ? * MON-FRI |
Weekdays at noon (Quartz) |
0 0 23 L * ? |
Last day of month at 23:00 (Quartz) |
0 0 9 15W * ? |
Nearest weekday to the 15th at 9:00 (Quartz) |
0 0 10 ? * 6#3 |
Third Friday at 10:00 (Quartz) |
0 30 10-14 ? * MON,WED,FRI |
Mon/Wed/Fri hourly 10:30–14:30 (Quartz) |
Troubleshooting: The Job Did Not Run
Validate the Expression, Then the Time Zone and Environment
- The expression: check the next fire time in a parser and confirm day/weekday are not misused.
- Time zone: Linux cron uses the server's local zone; Quartz uses the JVM zone; Kubernetes CronJob is fixed to UTC. The 09:00 you meant may be 17:00 in a container.
- Environment: cron's
PATHis often only/usr/bin:/bin; use absolute paths. - Permissions and line endings: the
.shneedschmod +x; a file edited on Windows can throwbad interpreter. - Logs: cron does not print to a terminal, so redirect output explicitly; on Debian/Ubuntu look for
CRONin/var/log/syslog.
When the frequency is wrong, check the step syntax first:
*/5 * * * * # ✅ minutes 0, 5, 10 … 55
0/5 * * * * # ⚠️ equivalent on some systems, not all
0-59/5 * * * * # ✅ explicit range with step
Setting the time zone in Quartz:
CronTrigger trigger = TriggerBuilder.newTrigger()
.withSchedule(CronScheduleBuilder
.cronSchedule("0 0 9 * * ?")
.inTimeZone(TimeZone.getTimeZone("Asia/Shanghai")))
.build();
Multiple Instances: Every Machine Runs the Same Expression
Standard cron has single-host semantics. Deploy several instances and each one fires the job, duplicating work. Two ways to converge:
- A distributed lock (Redis lock or a unique DB index) lets only one instance win.
- Move to a coordinator-aware scheduler (Quartz Cluster, xxl-job, or a K8s CronJob with
concurrencyPolicy: Forbid).
A lock only lowers the probability of overlap; the job still has to be idempotent, or retries and backfills will write duplicate data anyway.
Which Library Each Platform Uses
| Language / platform | Common library | Notes |
|---|---|---|
| Node.js | node-cron, cron |
5/6 fields; no L/W/# |
| Python | APScheduler, croniter |
croniter evaluates expressions only, it does not schedule |
| Java | Quartz, spring-task |
Quartz supports 6/7 fields and L/W/# |
| Go | robfig/cron |
5 fields by default; seconds optional |
| Kubernetes | CronJob |
Standard 5 fields, always UTC |
Reproducible Results
Two checks before shipping:
crontab -l # list this user's jobs and confirm the entry exists
grep CRON /var/log/syslog # Debian/Ubuntu: read the cron daemon's execution log
0 9 * * 1-5 means "Monday to Friday at 09:00". Paste it into the cron parser to see the field breakdown and the next fire time instantly. If the parse matches your intent but the log has no record at that timestamp, the cause is usually the environment or permissions rather than the expression.