dmesg -T results in inconsistent timestamp - linux

This question is not related to the pause/resume issue!
For the same event 'dmesg' shows always the same time stamp, e.g.
[31765279.760248]
However when using 'dmesg -T', for the same event it shows slightly different seconds count, e.g. calling dmesg -T | grep something | tail -1 twice results in:
[Thu Jan 29 01:12:39 2015] event details...
[Thu Jan 29 01:12:38 2015] event details...
I need to compare events with history and act on new events. However this difference makes it impossible to use a simple string compare.
As a work around I clip the seconds from the strings before I compare them. I can do it in this case since the events I filter do not happen more than once in 5 minutes.
Does anyone know why this inaccuracy happens?
uname -a => Linux (hostname) 3.5.0-45-generic #68~precise1-Ubuntu SMP Wed Dec 4 16:18:46 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

Those timestamps are converted to real time by adding them to the system boot time. If the system came up at 1491516481.480856282 and you add a fractional offset to it, sometimes you'll see one extra full second.
Haven't checked the code but pretty certain that's the reason. There's a bit more information in Convert dmesg timestamp to custom date format including a response I wrote.

Related

Format ps STIME on Linux

I recently moved to Linux from HP-UX and I noticed that STIME of a process is in "MonthDate" format(e.g. Apr21) while on HP-UX it was "Month Date"(e.g. Apr 21). There is no space between month and date. Is there a way to get a space between month and date?
HP-UX:
root 16773 1 1 Jul 15 ? 67:23 /opt/ssh/sbin/sshd
Linux:
oper 24494 23075 0 Apr21 ? 00:00:00 sshd: oper#pts/8
So once again I am digging procps-ng sources for the same function pr_stime() sources. You may be interested in the command above that function.
No, you can't parse stime consistently. The format is not specified in the standard. The output from procps-ng/ps varies depending how long ago the process was created. Procps-ng/ps uses 3 format strings depending on how long time ago the process has started. If your script depends on it, you may be surprised they all will stop working on new years eve when ps starts using %Y format string for stime column.
For determining how long time ago a process was started on linux, use etimes or etime with procps-ng/ps.

GNU date command not allowing for far away years?

When running the following command with GNU date :
date -d "20145-01-23"
I get the following result :
date: invalid date `20145-01-23'
However, the manual states that year can be any number :
For numeric months, the ISO 8601 format ‘year-month-day’ is allowed, where year is any positive number [...]
Thus, two questions :
is this a bug ? or am I misunderstanding something ?
how to obtain the same behavior than BSD date, which accepts this as a valid date ?
This works for me, using GNU coreutils 8.24:
$ date -d "20145-01-23"
Sat Jan 23 00:00:00 PST 20145
I'm not certain of this explanation (UPDATE: looks like my guess was correct), but my guess is that you're on a system with a 32-bit time_t rather than 64 bits. With 32 bits, the only representable dates are from Fri 1901-12-13 20:45:52 UTC (231 seconds before the epoch) to Tue 2038-01-19 03:14:07 UTC (231-1 seconds after the epoch).
If this is the case, then this command should succeed:
$ date -d 2037-01-01
Thu Jan 1 00:00:00 PST 2037
and this one should fail:
$ date -d 2039-01-01
date: invalid date ‘2039-01-01’
(I've reconstructed what the result should look like for that last command. The output of the first will vary slightly depending on your time zone.)
how to obtain the same behavior than BSD date, which accepts this as a valid date ?
Are you saying you want BSD date to reject it? If so (more speculation), either run a 32-bit version of BSD, or test whether the requested time is outside the 32-bit range of timestamps.
BSD date -d expects a time zone value. To offset years, days, or whatever, you pass -vNT, where N is a number and T is a char to indicate days, months, or years.
The flags differ. See http://www.unix.com/man-page/All/1/date/ for the Gnu version, http://www.unix.com/man-page/FreeBSD/1/date/ for a common BSD variant.

Get time in linux without DST and with DST

I want to know that is there any command which can provide time without DST if DST is applicable in the zone.
I have searched lot in google but not getting proper answer. I think there should be simple solution to get it.
Below is one link on stackoverflow.com but I am not getting
https://unix.stackexchange.com/questions/123493/disable-daylight-saving-time-in-debian-linux
For example:
current time in Newyork is
date
Wed Mar 23 04:51:54 EDT 2016
As per DST-free timezone definitions provided which just define the GMT-offset, called Etc/GMT±X:
TZ=Etc/GMT-1 date
Wed Mar 23 10:13:09 GMT-1 2016
Whereas DST is 1 hour forward on March 23 i.e. it should be ‎Wed, ‎Mar ‎23, ‎2016, ‏‎4:13 AM
Please anyone provide help.
U.S. Eastern time is often represented by the time zone name "America/New_York" or "US/Eastern". Equivalently, there is the "Posix" time zone name "EST5EDT". The essential fact here is that this zone is nominally 5 hours off of UTC (or 4 hours when daylight saving time is in effect).
There are also some DST-free zone names of the form "UTC-4" and "UTC+5".
So if you say
export TZ="UTC+5"
date
You'll see the date in the equivalent of U.S. Eastern Standard Time, without a DST correction.
(This is essentially what the high-rated answer at https://unix.stackexchange.com/questions/123493/disable-daylight-saving-time-in-debian-linux was trying to tell you, I think.)
If you wanted to take an arbitrary time zone name and construct from it an equivalent DST-free zone name, that'd be pretty tricky.

Why does the php 'date' function returns a wrong time (off by ~24 seconds)?

I have the following small php snippet running on a gentoo Linux (php version 5.2.10-pl0-gentoo):
#!/usr/bin/php5
<?
class TestDaemon {
public function __construct(){
while (TRUE){
unset($aDate);
exec("date", $aDate);
print("date(\"d.m.y H:i:s\") yields: ".date("d.m.y H:i:s")." while 'date' yields $aDate[0].\n");
sleep(1);
}
}
}
$oDaemon = new TestDaemon();
?>
And the output produced is as follows:
date("d.m.y H:i:s") yields: 27.03.14 07:05:27 while 'date' yields Thu Mar 27 07:05:03 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:28 while 'date' yields Thu Mar 27 07:05:04 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:29 while 'date' yields Thu Mar 27 07:05:05 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:30 while 'date' yields Thu Mar 27 07:05:06 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:32 while 'date' yields Thu Mar 27 07:05:07 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:33 while 'date' yields Thu Mar 27 07:05:09 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:34 while 'date' yields Thu Mar 27 07:05:10 UTC 2014.
date("d.m.y H:i:s") yields: 27.03.14 07:05:35 while 'date' yields Thu Mar 27 07:05:11 UTC 2014.
As you can see the times are off by approx. 24 seconds. On a different machine (same OS, same version of PHP) I do not see such an offset.
What is the reason for this offset? Does this come from leap second differences? Then which system gives the correct time? Why does php not use the Linux system time instead?
Also, can this time offset be a source of problems when working with the mysql database on the same system?
This looks like the shell running /bin/date is configured to use the "right" timezones and php is configured to use the POSIX-conformant timezones. The difference now should be 25 seconds, but if the tz data is over two years old then it would be 24 seconds. For a picture of why visit http://www.ucolick.org/~sla/leapsecs/amsci.html and see the second plot. The "right" zones follow the green line. The POSIX zones are required to stop the system clock on every leap second, so they follow the descending staircase of the blue line. [edit to be sure which was using which method]
Why the offset I don't know, but it seems that php date() uses time() function to get a timestamp. time() is defined like this:
Returns the current time measured in the number of seconds since the Unix Epoch (January 1 1970 00:00:00 GMT).
it never says in documentation that it uses System time.
According to this SO post the function time() uses the date.timezone set in php.ini or date_default_timezone_set().
So if your system uses a different time zone than your php.ini it could explain the difference because php seems to use its own time.
He also propose the following solution to get the real system time if they are not the same:
I'm going to give you a solution that works for Linux, I don't know for Windows. In Linux the system timezone is set in /etc/timezone.
Now, this is normally outside my allowed open_basedir setting, but you can add :/etc/timezone to your list to be able to read the file.
Then, on top of the scripts, that want to get the system time, you can call a library function that sets the script timezone to the system timezone. I suppose that this function is part of a class, so I use static:
static function setSystemTz() {
$systemTz = trim(file_get_contents("/etc/timezone"));
if ($systemTz == 'Etc/UTC') $systemTz = 'UTC';
date_default_timezone_set($systemTz);
}
To make the matter worse in PHP 5.3.3 'Etc/UTC' is not recognized, while 'UTC' is, so I had to add an if to fix that.
Not really answering your question but there are many different types of time variable on a typical Linux system:
The internal time of the harware clock (aka RTC or BIOS time), if you have one. This keeps the time when the system is offline, this may be fake and return bad values on a virtual machine, or you might not even have one. See hwclock
The amount of time that has passed since the kernel was started. (cat /proc/uptime)
The external time as provided by an NTP server via ntpd.
The innacuracy of the realtime clock / time source, how much it drifts over time in parts per million between samples. This is set by calls to the kernel. On a system with NTP it will be saved to your drift file (/var/lib/ntp/drift/ntp.drift or similar) and the system's time will be smoothly adjusted rather than have time jumps in your logs.
The number of miliseconds since 1st of January 1970. See date +%s
Your timezone, set in /etc/localtime, see zdump /etc/localtime
The current system time, calculated using all of the above plus rules for timezone, leap year, leap second and so on. See date
If I had to guess, I'd say that NTP is slowly adjusting your system time to compensate for your skewed realtime clock while PHP bypasses this and snags it from somewhere else.

How to store file creation time in linux

Is there any way to store file creation time in linux? May be filesystem that supports that. I mean CREATION TIME not mtime or ctime.
Most (all?) Linux filesystems do not store the time when the file was created. Instead, they just store when the file was last modified, when it was last accessed, and when it last had an administrative action applied to it (e.g., changing its permissions). Of those, the best approximation to what you're looking for is usually the modification time.
I do believe that Ext4 has creation time support.
Try this debugfs:
debugfs -R 'stat <inode>' /dev/block_dev.
Fragment: Address: 0 Number: 0 Size: 0
ctime: 0x51872d50:df6167a0 -- Mon May 6 07:10:56 2013
atime: 0x5183b4e1:d2ad63cc -- Fri May 3 16:00:17 2013
mtime: 0x51872d50:df6167a0 -- Mon May 6 07:10:56 2013
crtime: 0x5183b4e1:d2ad63cc -- Fri May 3 16:00:17 2013
Understanding EXT4 (Part 2): Timestamps
On NTFS using ntfs-3g you can access the creation time via an extended attribute (see attr(5)).
If you want to store file creation times, you can use filesystem event notification tool like inotify.
It has system calls that make you aware of IN_CREATE (a file creation event) or some other event, When a file creation event happens you can simply log that with current time in a file of your preference.

Resources