[chrony-users] time.google.com crash chronyd

[Nicolas Embriz]

I am testing chronyd 3.1 on FreeBSD 11.1 with this config:


server time.google.com
user chronyd
driftfile /var/db/chrony/drift
dumponexit
dumpdir /var/db/chrony
pidfile /var/run/chronyd.pid
stratumweight 0
makestep 1.0 3
logchange 0.5
bindcmdaddress 127.0.0.1
bindcmdaddress ::1
allow

For an unknown reasons when using time.google.com chronyd don’t stat but if
I use for example “time.apple.com” it works, this is the output I get when
using chronyd -d:


2018-05-06T11:51:41Z chronyd version 3.1 starting (+CMDMON +NTP +REFCLOCK
-RTC +PRIVDROP -SCFILTER +SECHASH -SIGND +ASYNCDNS +IPV6 -DEBUG)
2018-05-06T11:51:41Z Frequency 0.090 +/- 0.037 ppm read from
/var/db/chrony/drift
Assertion failed: (MD5_hash >= 0), function UTI_IPToRefid, file util.c,
line 373.
Abort

Probably related to this:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=223840

Any ideas about how to make it work?

Thanks in advance.

Regards.

[Alvaro G. M.]

For an unknown reasons when using time.google.com chronyd don’t stat but if
I use for example “time.apple.com” it works, this is the output I get when

I can't help you with that, but I think you should be aware that
time.google.com behaviour is non standard, so you may not want to use it.

https://github.com/systemd/systemd/issues/437

(You can ignore all the systemd flame, just the main post is enough and has
the references you may need to further investigate this thing).

Best luck!

[Bill Unruh]

Post by Alvaro G. M.

Post by Nicolas Embriz
For an unknown reasons when using time.google.com chronyd don’t stat but if
I use for example “time.apple.com” it works, this is the output I get when

I can't help you with that, but I think you should be aware that
time.google.com behaviour is non standard, so you may not want to use it.
https://github.com/systemd/systemd/issues/437

Just to be clear, google treats leapseconds in a non-standard way. ntpd and
chrony step the clock over a leapsecond-- either forward or backwark at the
leapsecond. Google however smears the leapsecond before and after the
leapsecond time, over about an hour. If something sets the leapsecond flag,
then google time will start smearing the leapsecond an hour before (eg,
speeding up their clocks by about 150PPM) an hour before the leapsecond should
occur, and up to an hour afterwards, but without putting in a leapsecond. Now
if other time source sets the leapsecond flag on your chrony, then there is a
danger that the leapsecond will also be applied as a step by chrony.
completely messing things up around the leapsecond time.

So, don't use google. Use pool.ntp.org or get yourself a timeing gps (gps with
PPS)and set your system up as a stratum 1 time using gps with pps if correct
timing is important to you. (pool, since it uses volunteered servers, can be a
less than exact in its time delivery-- you should not expect better than
millisecond accurcy from the pool)

Post by Alvaro G. M.
(You can ignore all the systemd flame, just the main post is enough and has
the references you may need to further investigate this thing).

Well, since it is very likely that the OP got his default from systemd, it is
not clear that the systemd comments are inappropriate. I certainly thing it is
inappropriate for systemd to use google as their default. Users are going to
read "default" as "recommended"-- why would anyone be so irresponsible as to
use a default which is defective. Of course systemd may have altered this
since 2015.
Nope. From timesyncd.conf
#FallbackNTP=time1.google.com time2.google.com time3.google.com time4.google.com
It is commented out, but this certainly is silly. (It might be mageia that
commented it out, or it may come directly from systemd)
And in man timesyncd.conf
"If this option is not given, a compiled-in list of NTP servers is used
instead"

And I suspect it is google's server that is compiled in. This is really
irresponsible.

So, the question is -- where did the OP get his "timeX.google.com" from that
he is using.

Post by Alvaro G. M.
Best luck!
--
Alvaro G. M.
--
with "unsubscribe" in the subject.
with "help" in the subject.

[Bill Unruh]

Just to amplify what I said, here is a link to google's page
https://developers.google.com/time/smear
They use a one day smear rather than the one hour smear I suggested in my
response.

Note that this should ONLY be a problem around a leapsecond, and there has
been no leapsecond for a while (Dec 31,2016 was the last one), so it is highly doubtful that the OP's
problems with time.google.com originated from this smearing of the time around
a leapsecond.

Post by Alvaro G. M.

Post by Nicolas Embriz
For an unknown reasons when using time.google.com chronyd don’t stat but if
I use for example “time.apple.com” it works, this is the output I get when

I can't help you with that, but I think you should be aware that
time.google.com behaviour is non standard, so you may not want to use it.
https://github.com/systemd/systemd/issues/437
(You can ignore all the systemd flame, just the main post is enough and has
the references you may need to further investigate this thing).
Best luck!
--
Alvaro G. M.
--
with "unsubscribe" in the subject.
with "help" in the subject.