The editorial frames the no-leap-second announcement as the ninth consecutive quiet Bulletin C, arguing the pause isn't about better clocks — it's because Earth has been spinning faster over the past decade. Citing IERS length-of-day data and Duncan Agnew's 2024 Nature paper on ice-melt mass redistribution, it notes UT1 is now drifting ahead of UTC rather than behind, inverting the assumption that leap seconds were built on.
The editorial calls leap seconds the industry's 'dirty secret,' cataloging major outages: the 2012 insertion crashed Reddit, LinkedIn, Yelp, and Qantas via a Linux hrtimer deadlock, and Cloudflare's DNS resolver broke in 2017 on negative time deltas. The argument is that a mechanism designed to keep civil time aligned with Earth rotation has repeatedly proven more dangerous than the drift it corrects.
The editorial argues UTC's compromise — atomic SI seconds punctuated by occasional 61-second minutes to stay within 0.9s of UT1 — only worked while Earth was reliably slowing down. Now that rotation is accelerating, the same mechanism may soon require a negative leap second, a scenario most software has never been tested against, exposing UTC as a design built on an assumption that no longer holds.
By submitting the raw IERS Bulletin C directly to Hacker News and drawing 259 points and 195 comments, the submitter treats the non-event as material news for operators. The framing implicit in surfacing the bulletin is that 'no leap second' is precisely the outcome the infrastructure community wants to hear confirmed on schedule.
The International Earth Rotation and Reference Systems Service (IERS) published Bulletin C on schedule, and the answer is the one most sysadmins wanted: no leap second will be introduced at the end of December 2026. The difference between UTC and TAI stays fixed at 37 seconds, exactly where it's been since January 1, 2017.
That's now nine consecutive Bulletin C releases with no insertion — the longest quiet stretch since leap seconds were introduced in 1972. Between 1972 and 1999, IERS inserted 22 of them, roughly one every 18 months. Since 2000, only five. Since 2017, zero.
The reason isn't that timekeeping got easier — it's that Earth started spinning faster. Length-of-day measurements from IERS show the planet has shaved fractions of a millisecond off the mean solar day over the past decade, driven by a mix of core-mantle coupling, glacial rebound, and — according to a 2024 Nature paper by Duncan Agnew — ice-melt-driven mass redistribution partially offsetting the acceleration. The net effect: UT1 (Earth-rotation time) is drifting *ahead* of UTC, not behind it.
Leap seconds exist because two definitions of "a second" disagree. The SI second is defined by cesium-133 atomic transitions and doesn't care what Earth does. UT1 is tied to Earth's actual rotation, which wobbles. UTC splits the difference by running on atomic seconds but occasionally inserting a 61-second minute so it stays within 0.9s of UT1. That worked fine when the planet was reliably slowing down. It works badly now that it isn't.
The industry's dirty secret is that leap seconds have caused more outages than they've ever prevented. The 2012 insertion took down Reddit, LinkedIn, Yelp, and Qantas's check-in system when the Linux kernel's `hrtimer` subsystem deadlocked on the repeated 23:59:59. Cloudflare's 2017 leap second broke its DNS resolver for about an hour when negative time deltas fell through unhandled code paths. Google infamously stopped participating in the ceremony entirely, adopting a "leap smear" that spreads the extra second across 24 hours of slightly-longer atomic seconds. AWS did the same in 2015. Meta joined the anti-leap-second campaign publicly in 2022, calling for abolition.
They got their wish, sort of. In November 2022, the CGPM (Conférence Générale des Poids et Mesures) voted to abolish leap seconds by 2035, replacing them with a much larger tolerance — probably a leap minute or leap hour, decided decades from now when the drift becomes visible. But 2035 is still nine years off, and the more immediate risk is the mirror-image problem: a negative leap second.
If Earth keeps speeding up, IERS will eventually have to *remove* a second to keep UTC synced with UT1. That's never happened. Every leap-second bug in production code was written by someone who assumed the correction would be positive. Judah Levine of NIST has warned repeatedly that a negative leap second — turning 23:59:58 directly into 00:00:00, skipping 23:59:59 — would surface a class of bugs the industry has never stress-tested. Timestamps that go backward inside a monotonic-clock assumption. Cron jobs scheduled for a second that never happens. Kafka offsets, database WALs, distributed consensus protocols — anything that treats time as monotonically increasing at exactly 1s/s.
The good news buried in Bulletin C is that we get another year of runway before that experiment becomes real.
If you're running anything that touches wall-clock time — which is most things — the Bulletin C announcement is a nudge to do the audit you've been postponing.
Check what your clock source actually does during a leap event. If you're on AWS, GCP, or Meta infra, you're already on a smeared clock and probably fine. If you're running your own NTP infrastructure with `chrony` or `ntpd`, know which policy is configured — `leapsectz`, slew, step, or smear — and confirm your kernel's `CLOCK_TAI` behavior matches your assumption. Public NTP pool servers still deliver the raw leap-second flag; your OS decides what to do with it.
Prefer monotonic clocks for durations. `clock_gettime(CLOCK_MONOTONIC)` on Linux, `mach_absolute_time` on Darwin, `QueryPerformanceCounter` on Windows. Any code that computes an elapsed interval by subtracting two `Date.now()` calls is a bug waiting for the next leap event — positive or negative.
Grep your codebase for `:60` and `:61`. ISO 8601 technically allows `23:59:60` for a positive leap second. A lot of parsers don't. A lot of *your* parsers probably don't. Test them. And when negative leap seconds arrive, the failure mode flips: a timestamp your regex accepts today (`23:59:59`) simply won't exist on the day it happens, and any code that polls at exactly that second will silently miss its window.
The 2035 abolition vote is the real headline, but it's a decade of half-measures away. Between now and then, the working assumption should be: positive leap seconds are probably done, negative leap seconds are probably coming, and the code you ship in 2027 will still be running when either happens. Bulletin C bought us twelve more months of the boring outcome. Use them to make sure the exciting one — whenever it lands — is a non-event.
Isn't this anti-news? I vaguely remember they have stopped introducing leap seconds until further notice because they cause too much trouble with today's computing systems. On my phone now and did not research it, as I say vaguely remember. Now we are 37 seconds off. Nobody of us will have
ELI5: How does this impact UNIX timestamps? Particularly for things that are in maintenance mode or otherwise minimally maintained.Nothing I do requires this level of precision, but certainly there are things that do.
"To authorities responsible for the measurement and distribution of time" is just the best preamble ever.
If the UTC-TAI offset remains at -37s, then it also means the UTC-GPS offset remains at -18s. TAI and GPS have a constant 19s offset from each system.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
What causes the unpredictability in this? I would have guessed we have earth's rotation and orbit down to many decimals. Does geological activity, weather, or something else cause rotation speed differences that we just can't predict?