strconv.ParseInt: parsing "18446744073709551615": value out of range

[gdejong]

After upgrading prometheus/procfs from 0.6.0 to 0.7.0 the process_cpu_seconds_total, process_virtual_memory_bytes and process_resident_memory_bytes metrics from prometheus/client_golang were no longer outputted.

We have a Go service running in a Docker container.

Using the ReportErrors option from ProcessCollectorOpts I was able to display the error.

prometheus.MustRegister(prometheus.NewProcessCollector(prometheus.ProcessCollectorOpts{ReportErrors: true, Namespace: "test_"}))

Scraping the metrics gave:

An error has occurred while serving metrics:

strconv.ParseInt: parsing "18446744073709551615": value out of range

This is the maximum value of a uint64

The /proc/1/stat file contents:

# cat /proc/1/stat
1 (service) S 0 1 1 0 -1 1077944576 4345 0 1 0 13 17 0 0 20 0 5 0 346180303 732237824 2193 18446744073709551615 4194304 9876697 140737276107232 140737276105984 4665027 0 0 0 2143420159 18446744073709551615 0 0 17 4 0 0 1 0 0 14888960 15220704 24559616 140737276112712 140737276112721 140737276112721 140737276112879 0

This is parsed in proc_stat.go -> Stat()

The value 18446744073709551615 occurs twice. On position 23 and 33.
Position 33 is written into ignored which is defined as an int. A value as large as 18446744073709551615 cannot be read into an int datatype and this causes the error: strconv.ParseInt: parsing "18446744073709551615": value out of range

Possible solution: change the datatype of ignored from int to uint64?

[gdejong]

Some extra information.

The /proc/<pid>/stat format is defined here: https://man7.org/linux/man-pages/man5/proc.5.html

All fields are listed there with their proper scanf format (https://man7.org/linux/man-pages/man3/scanf.3.html)

The 25th and 35th element are defined as:

              (25) rsslim  %lu
                     Current soft limit in bytes on the rss of the
                     process; see the description of RLIMIT_RSS in
                     getrlimit(2).
---

              (35) wchan  %lu  [PT]
                     This is the "channel" in which the process is
                     waiting.  It is the address of a location in the
                     kernel where the process is sleeping.  The
                     corresponding symbolic name can be found in
                     /proc/[pid]/wchan.

l - Indicates either that the conversion will be one of d, i, o, u, x, X, or n and the next pointer is a pointer to a long or unsigned long (rather than int)
u - Matches an unsigned decimal integer; the next pointer must be a pointer to unsigned int.

So lu would become an unsigned long

This would lead me to assume that the proposed solution would be correct.

[vykulakov]

Noticed the same issue with parsing /proc/[pid]/stat on some machines:

strconv.ParseInt: parsing \"18446744073709551615\": value out of range

It happened on parsing the following stat file:

29977 (test)  S 15396 29977 15396 34818  29977 1077944320 4995      0         0    0    77      241    0      0      20 0  24 0  3933239738 1603444736 2552  18446744073709551615 4194304 8586288  140734181603184 140734181602680 465270 7  0  1006254592 0  2143420159 18446744073709551615 0  0  17 17 0  0  0  0  0  12419072 12726432 42594304 140734181607380 140734181607391 140734181607391 140734181609453 0

The code that I used to build that test program is this (the last error check fires):

	pid := os.Getpid()
	proc, err := procfs.NewProc(pid)
	if err != nil {
		logrus.WithField("pid", pid).WithError(err).Error("cannot get proc for the current process")
		return
	}

	stat, err := proc.Stat()
	if err != nil {
		logrus.WithField("pid", pid).WithError(err).Error("cannot get proc stat for the current process")
		return
	}

The test program is run on the following kernel:

# uname -a
Linux example.com 3.10.0-1062.18.1.el7.x86_64 #1 SMP Tue Mar 17 23:49:17 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

On kernels with version 5.10 all works fine (what a relief).

[vykulakov]

But we cannot just replace the int type for the ignored variable with the uint64 type because this variable is also used for the field 21:

              (21) itrealvalue  %ld

that means signed long. The code should have different ignored variables for different fields, I guess.

[vykulakov]

I'll prepare PR.

[gdejong]

I agree, multiple ignored variables with different datatypes would be appropriate.

[vykulakov]

@gdejong thank you for the links right here.

Please, review the PR and maybe it will be merged soon. Thanks.