About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
timedatectl printing System clock synchronized: no means no network time source is currently disciplining the clock. On a server that runs chrony, two commands find the reason quickly: chronyc tracking (is the clock synchronised, and to what) and chronyc sources (can chronyd reach any server at all). Causes include blocked UDP port 123, a server name that does not resolve, a clock so far out that chrony will only slew it, and no time service running.
What the symptom means
timedatectl "may be used to query and change the system clock and its settings, and enable or disable time synchronization services." The systemd manual's example output for a healthy machine shows the lines System clock synchronized: yes and NTP service: active. When the first line says no, the clock is not being synchronised at that moment. The NTP service line tells you whether a service that should do it is active.
Chrony does not jump the clock by default. The chronyc manual says that "normally chronyd will cause the system to gradually correct any time offset, by slowing down or speeding up the clock as required." It will step only if you allow it with the makestep directive or run the makestep command. And it only counts a source once the source answers.
The chronyc tracking output includes a field called Leap status, which the manual says "can be Normal, Insert second, Delete second or Not synchronised." The Reference ID field names the server the computer is synchronised to. If it is 7F7F0101 with no name or address, the manual says the computer "is not synchronised to any external source" and chrony is in local mode.
Likely causes
| Cause | What you see | How to confirm |
|---|---|---|
| No time service is running | NTP service: inactive |
systemctl is-active on the chrony and timesyncd units; chronyc fails with "506 Cannot talk to daemon" |
| Firewall blocks UDP 123 | chronyc sources shows Reach 0 for every source |
tcpdump on UDP port 123 shows requests but no replies |
| Server names do not resolve | chronyc activity reports sources with unknown address |
chronyc -N sources -a lists names with no address |
| Clock too far out to slew | chronyc tracking shows a large System time that falls slowly |
Compare System time with the makestep setting in chrony.conf |
| Source rejected as inaccurate | chronyc sources shows a reachable source marked ?, x or ~ |
chronyc ntpdata shows large root dispersion, or x marks a falseticker |
| A different time service is active | timedatectl and chronyc disagree |
Both chronyd and systemd-timesyncd show as active |
Prove the cause before changing anything
These commands read only.
timedatectl
chronyc tracking
chronyc sources -v
chronyc activity
systemctl is-active chronyd.service systemd-timesyncd.service
Unit names vary by distribution; the chrony FAQ names chronyd.service, and the systemd manual names systemd-timesyncd.service. Use your distribution's name for the chrony service if it differs.
Read chronyc sources -v, which adds caption lines. The manual defines the columns. The first character is the mode (^ is a server). The second is the selection state: * is the source currently selected for synchronisation, + is combined with it, - is selectable but not selected, x is "a falseticker", meaning its time is inconsistent with a majority of the other sources, ~ is "too much variability", and ? is "not considered to be selectable for synchronisation for other reasons (e.g. unreachable, not synchronised, or does not have enough measurements)".
Reach is "the source's reachability register printed as an octal number." A value of 377 means a valid response to each of the last 8 polls. The chrony FAQ says that if Reach "is zero, it means chronyd did not get any valid responses from the NTP server" and that a firewall may block requests to UDP port 123 of the server or the replies. It suggests Wireshark or tcpdump to see if responses arrive. A capture filter for this is udp port 123, using the pcap-filter syntax port together with udp:
tcpdump -n -i eth0 udp port 123
-n stops name conversion and -i selects the interface, per the tcpdump manual. Replace eth0 with your interface.
For name problems, the FAQ says that a non-zero count of "sources with unknown address" in chronyc activity points at name resolution, and from chrony 4.0 chronyc -N sources -a prints all sources including those without a known address. The FAQ adds a trap: with DNSSEC on, name resolution fails when the clock is far from true time, so chronyd cannot resolve its servers. For that case it says to use bare IP addresses or to disable DNSSEC validation for chronyd.
If you suspect a Windows time server, the FAQ says these "report a very large root dispersion (e.g. three seconds or more)", so chronyd ignores them. chronyc ntpdata shows it. The FAQ's fix is a larger maxdistance, for example maxdistance 16.0.
Example (illustrative): app01 runs chrony with pool pool.ntp.org iburst and makestep 1 3. After a firewall change, its clock is 90 seconds fast and timedatectl says System clock synchronized: no. chronyc sources lists four servers with Reach 0. tcpdump -n -i eth0 udp port 123 shows outgoing requests and no replies, so the requests leave the server and nothing comes back. The administrator looks for a rule that blocks UDP 123 on the path to the time servers or on the return path, finds the one the firewall change added, and removes it. Reach then climbs towards 377. The makestep 1 3 directive steps the clock if the offset is above 1 second, but only if there were no more clock updates since chronyd was started than the limit of 3 (chrony.conf manual). app01 had been running and synchronised for weeks before the change, so its three updates were used long ago and chronyd does not step: it slews the 90 seconds, which takes a long time. To correct it at once, the administrator runs chronyc makestep in a maintenance window (see the Risk under "Clock far out: allow a step"). A restart of chronyd would also reset the update count, which makes the directive apply again.
Fix
No service running
timedatectl set-ntp true "enables and starts the first existing network synchronization service." Which services count is set by .list files in ntp-units.d/ directories, or by the SYSTEMD_TIMEDATED_NTP_SERVICES variable. The systemd example for choosing chrony is SYSTEMD_TIMEDATED_NTP_SERVICES=chronyd.service:systemd-timesyncd.service. The systemd manuals do not say what happens if two services run together. Pick one and keep the other stopped.
Disruptive step: set-ntp true or set-ntp false enables, starts, disables and stops services. Back out with the opposite value.
Blocked UDP 123 or DNS
Allow outbound UDP 123 from the server to the time sources, and make sure the resolver in /etc/resolv.conf works. The FAQ documents both causes.
Risk: a firewall change affects every connection that the rule covers, not only time traffic. A rule that is too wide opens more than the time sources need, and a rule that replaces an existing one can cut your own access. Keep the change as narrow as the rule that was missing: UDP 123 to the time sources only. Record the rule set before the change, keep a console open, and back out by restoring the earlier rule set.
Clock far out: allow a step
The FAQ's example makestep 1 3 steps the clock "in the first three updates if its offset was larger than one second", and says that on virtual machines that can be suspended and resumed with an incorrect time, you might allow stepping on any update with makestep 1 -1. The chrony.conf manual says the directive applies "only if there were no more clock updates since chronyd was started than the specified limit", so on a daemon that has been running for a long time a limit of 3 no longer applies.
Risk: editing the makestep line in chrony.conf changes when chrony may jump the clock. The file is read when chronyd starts, so the change takes effect after sudo systemctl restart chronyd.service, which also restarts the daemon's measurements and can step the clock on the first updates. The chronyc manual also documents a form that changes the automatic stepping while chronyd runs, chronyc makestep threshold limit (for example chronyc makestep 1 3), which "configures the automatic stepping, similarly to the makestep directive". That form is not saved in the file. To back out, restore the earlier line and restart chronyd. Time that has already been stepped cannot be undone. The chronyc manual also offers a one-time command, chronyc makestep, which "tells chronyd to cancel any remaining correction that was being slewed and jump the system clock by the equivalent amount."
Risk: a step changes the system time instantly. The manual warns: "Certain software will be seriously affected by such jumps in the system time." Databases, Kerberos tickets, log ordering, certificates and schedulers can all notice. Do it in a window and, where possible, stop time-sensitive services first. There is no way back to the old time; you can only step again.
Slow startup ordering
The chronyc manual describes chronyc waitsync, which waits until chronyd has synchronised, with an example of waitsync 60 0.01 that waits up to about 10 minutes for a correction below 10 milliseconds. Use it in scripts that must not start before the clock is right.
chronyc cannot talk to the daemon
The FAQ's "506 Cannot talk to daemon" answer is to check that chronyd is running and that UDP port 323 is not blocked for remote chronyc use. Local chronyc uses a Unix socket and needs no firewall rule.
Confirm the fix
timedatectlshowsSystem clock synchronized: yesandNTP service: active.chronyc trackingshowsLeap status : Normal, aReference IDnaming a real server, and a smallSystem time.chronyc sourcesshows one*source and Reach at or near377.
What to watch afterwards
- Reach and the
*source after firewall changes, because UDP 123 rules are easy to lose. System timeandLast offsetinchronyc trackingafter VM suspends, snapshots and migrations.- A second time service enabled by an image or a package update.
- Time-dependent systems downstream. The page titled "Clock skew between identity and application hosts in production" explains why a few minutes matters to Kerberos and token validation.
Read next
Related articles you may want next.
- HostingClock skew between identity and application hosts in productionHow Kerberos and token time windows turn small clock drift between identity and application servers into production auth failures—and how to detect skew early.Read article
- Backup & RecoveryS3 backup uploads fail with RequestTimeTooSkewed: the server clock, not the bucketS3 backups fail with RequestTimeTooSkewed when the host clock is wrong. Prove the offset with chrony, timedatectl or w32tm, then fix and verify.Read article
- VMwareESXi host time is wrong: NTP shows 'Time Synchronized: false' or 'Configuration is not working normally'ESXi host clock wrong and NTP not syncing? Read ntpq reach and refid, check the ntpClient firewall rule and UDP 123, then fix and confirm with esxcli.Read article
Risks and limits
- These commands have not been run or lab-tested. They are taken from the manuals and FAQ cited in Sources.
- The manuals describe chrony 4.9 and systemd 262. Older versions may lack options such as
chronyc -N(the FAQ places it at chrony 4.0 or later). - The systemd manuals do not state a rule about running chrony and timesyncd together. Running one service is advice, and the choice is yours to make; no manual cited forbids running both.
- Virtual machine host time-sync settings are outside these sources and can also move a guest clock.
- Debian-family file paths and unit names may differ from the names in the chrony documentation.
Sources
- chronyc(1) manual, chrony 4.9 — chrony project, page shows "Last updated 2026-08-27", accessed 18 November 2026
- chrony.conf(5) manual, chrony 4.9 — chrony project, page shows "Last updated 2026-08-27", accessed 18 November 2026
- chronyd(8) manual, chrony 4.9 — chrony project, page shows "Last updated 2026-08-27", accessed 18 November 2026
- chrony FAQ — chrony project, page shows "Last updated 2026-08-27", accessed 18 November 2026
- timedatectl(1) — systemd project, header "systemd 262", no date shown, accessed 18 November 2026
- systemd-timedated.service(8) — systemd project, no date shown, accessed 18 November 2026
- systemd-timesyncd.service(8) — systemd project, no date shown, accessed 18 November 2026
- systemctl(1) — systemd project, no date shown, accessed 18 November 2026
- tcpdump(1) (page shows "Updated: 10 September 2026") and pcap-filter(7) (page shows "Updated: 16 September 2026") — tcpdump project, accessed 18 November 2026
This article was originally published on .