We read /proc/stat using a statically sized buffer. This was increased
to 64kB in 2005 via commit 777fcd3cf15d "/proc/stat for 1024 CPUs".
Unfortunately in 2020, 1024 CPUs is not enough for anyone. I have a
large machine where /proc/stat is 74kB, and vmstat gives incorrect
output.
Double the buffer to 128kB in the confidence that 2048 CPUs is actually
enough for anyone.
Signed-off-by: Anton Blanchard <anton@ozlabs.org>
New command, pwait! Waits for another process to finish just like
pgrep finds or pkill kills another process.
References:
procps-ng/procps!97
Signed-off-by: Craig Small <csmall@dropbear.xyz>
I removed the old man pages but didn't remove their references in
the Makefile
References:
commit a978f68c06eb1764d401fa0ffce9fb30cd55f2e4
commit 770505ded42566b70fd0ee2cade98a317d18ac38
commit 7246249b8c2683c13515147ff4f9dbb85ba211ae
With glibc, each time the strftime() function is used (twice per process
in a typical ps -fe run), a stat("/etc/localtime") system call is used
to determine the timezone. Not only does this add extra system call
overhead, but when multiple ps processes are trying to access this
file (or multiple glibc programs using strftime) in parallel, this can
trigger significant lock contention within the OS kernel.
Since ps is not intended to run for long periods of time as a
daemon (during which the system timezone could be altered and PS might
reasonably be expected to adapt its output), there is no benefit to
repeatedly doing this stat(). To stop this behavior, explicitly set the
TZ variable to its default value (:/etc/localtime) whenever it is unset.
glibc will then cache the stat() result.
With glibc, each time the strftime() function is used (twice per process
in a typical ps -fe run), a stat("/etc/localtime") system call is used
to determine the timezone. Not only does this add extra system call
overhead, but when multiple ps processes are trying to access this
file (or multiple glibc programs using strftime) in parallel, this can
trigger significant lock contention within the OS kernel.
Since ps is not intended to run for long periods of time as a
daemon (during which the system timezone could be altered and PS might
reasonably be expected to adapt its output), there is no benefit to
repeatedly doing this stat(). To stop this behavior, explicitly set the
TZ variable to its default value (:/etc/localtime) whenever it is unset.
glibc will then cache the stat() result.
While sysctl did change the order of /run and /etc to match
systemd in the referenced commit, the Debian bug report that
brought it to light was not documented.
References:
commit 24a1574f0acbe636c98ab13b439c93f7b91de697
https://bugs.debian.org/950788
While sysctl did change the order of /run and /etc to match
systemd in the referenced commit, the Debian bug report that
brought it to light was not documented.
References:
commit 24a1574f0acbe636c98ab13b439c93f7b91de697
https://bugs.debian.org/950788
If a hash results report was output (via ATEOJ_RPTHSH)
a portion is devoted to occupied table entries ordered
by depth. There is a possibility that some depths will
not be found among existing occupied table entries and
to avoid any confusion probably should not be printed.
[ to illustrate the potential for confusion prior to ]
[ this patch, force a very small table size (like 8) ]
[ and then arrange to trigger the end-of-job report. ]
So this patch ensures only 'in use' entries are shown.
[ admittedly, all of the remaining logic in the loop ]
[ could/should be subordinate to this new 'if' test, ]
[ but we will keep the change to a minimum. besides, ]
[ there's no harm subtracting/adding a zero numdepth ]
[ especially since the chance of a zero is very low. ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
So as to not obscure the results from this new target,
we'll redirect that final 'make clean' output to null.
Signed-off-by: Jim Warner <james.warner@comcast.net>
If a hash results report is output (via UNREF_RPTHASH)
a portion is devoted to occupied table entries ordered
by depth. There is a possibility that some depths will
not be found among existing occupied table entries and
to avoid any confusion probably should not be printed.
[ to illustrate the potential for confusion prior to ]
[ this patch, force a very small table size (like 8) ]
[ & then trigger the procps_pids_unref() eoj report. ]
So this patch ensures only 'in use' entries are shown.
[ admittedly, all of the remaining logic in the loop ]
[ could/should be subordinate to this new 'if' test, ]
[ but we will keep the change to a minimum. besides, ]
[ there's no harm subtracting/adding a zero numdepth ]
[ especially since the chance of a zero is very low. ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
free, slabtop and uptime would happily take extra command line
arguments and doing nothing about them. The programs now check
optind after option processing and will give you usage screen
if there is anything extra.
References:
procps-ng/procps#181
free, slabtop and uptime would happily take extra command line
arguments and doing nothing about them. The programs now check
optind after option processing and will give you usage screen
if there is anything extra.
References:
procps-ng/procps#181
For long lines from a process, watch would wrap them around to the
next. While this default option has it uses, sometimes you want to
just cut those long lines down.
watch has a -w flag which will truncate the lines to the number
of columns. A few simple lines to do this new trick.
I think I caught all the ANSI state correctly but there might be
a chance it bleeds to the next row.
References:
procps-ng/procps#182
For long lines from a process, watch would wrap them around to the
next. While this default option has it uses, sometimes you want to
just cut those long lines down.
watch has a -w flag which will truncate the lines to the number
of columns. A few simple lines to do this new trick.
I think I caught all the ANSI state correctly but there might be
a chance it bleeds to the next row.
References:
procps-ng/procps#182
noinst_PROGRAMS are built with "make" even though we had the
test programs in there and only needed them for "make check".
In theory the check target should depend on check_PROGRAMS as
check-am target does and the document states it should, but for
reasons understood by the automake whisperers only, it doesn't
build them.
check only depends on BUILT_SOURCES for some reason.
check-am: all-am
$(MAKE) $(AM_MAKEFLAGS) $(check_PROGRAMS)
$(MAKE) $(AM_MAKEFLAGS) check-TESTS
check: $(BUILT_SOURCES)
$(MAKE) $(AM_MAKEFLAGS) check-recursive
References:
https://www.gnu.org/software/automake/manual/html_node/Scripts_002dbased-Testsuites.html
The referenced commits created the library infrastructure and test
program to validate that the structures and macros line up with
each other.
The library needs to be (re)built with -DITEMTABLE_DEBUG and then
the test program ran. We clean before and after so we are not
testing a non-debug library or having a debug library hanging around
to cause future problems.
Due to test_Itemtables depending on the library, we don't need to
explicitly build the library.
To validate the library structure/header corrospondence run:
make check-lib
References:
commit e616409aa497b5dc656dc7a5bd0a21adf0ee4d36
commit 92d0297e1e4d5946c5b098e37c91c7e524a0eca0
https://www.freelists.org/post/procps/keep-on-patchin,19
This patch just raises the size of the hash table used
to calculate elapsed task stuff. The net result should
be less need for 'chaining' under pid hash collisions.
[ the hash scheme is intentionally kept as primitive ]
[ and, therefore, as fast as possible. it employs an ]
[ 'and' approach versus a 'mod' operation since both ]
[ yield similar distribution but the former approach ]
[ was 4 fewer cpu instructions in terms of overhead. ]
[ additionally, for hash collisions, 'chaining' uses ]
[ an array index rather than the usual pointer since ]
[ the HST_t guys may move when they are reallocated. ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
This patch separates the memory allocations into those
used initially from those used in later reallocations.
Thus, we can reduce that iterative realloc() overhead.
Additionally, we'll correct a long standing oops where
multiple history_info structures were created at 'new'
time when only one should have been allocated (jeeze).
[ originally the allocation was strangely based upon ]
[ number of 'items' (???) & later a #define constant ]
Reference(s):
. May, 2016 - subsequent bad history_info logic
commit 9ebadc1438a6665a98a9f0782523b0f9a2a6248f
. Aug, 2015 - original faulty history_info code
commit 7e6a371d8a36b250a2edddff9f5d059640b8132e
Signed-off-by: Jim Warner <james.warner@comcast.net>
This patch just raises the size of the hash table used
to calculate elapsed task stuff. The net result should
be less need for 'chaining' under pid hash collisions.
[ the hash scheme is intentionally kept as primitive ]
[ and, therefore, as fast as possible. it employs an ]
[ 'and' approach versus a 'mod' operation since both ]
[ yield similar distribution but the former approach ]
[ was 4 fewer cpu instructions in terms of overhead. ]
[ additionally, for hash collisions, 'chaining' uses ]
[ an array index rather than the usual pointer since ]
[ the HST_t guys may move when they are reallocated. ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
The referenced commit the comm length was increased from 16 to 64
characters to handle the larger command names for things like kernel
threads.
However most user processes are limited to 15 characters which means
if you try something like ps -C myprogramisbiggerthansixteen this would
fail to match because /proc/<PID>/comm would only be myprogramisbigg
ps now checks the comm length and if it is 15 and if the given match
is 15 or more, it will only match the first 15 characters.
This is also how killall has worked for about a year.
Thanks to Jean Delvare <jdelvare@suse.de> for the note.
Copy of commit from master.
References:
commit 14005a371e5c14289e96a4927ffd1a827d3c9d85
commit psmisc/psmisc@1188315cd0
commit 3e1c00d051ae4a8319df7714d0545b4b7aa81557
Signed-off-by: Craig Small <csmall@dropbear.xyz>
The referenced commit the comm length was increased from 16 to 64
characters to handle the larger command names for things like kernel
threads.
However most user processes are limited to 15 characters which means
if you try something like ps -C myprogramisbiggerthansixteen this would
fail to match because /proc/<PID>/comm would only be myprogramisbigg
ps now checks the comm length and if it is 15 and if the given match
is 15 or more, it will only match the first 15 characters.
This is also how killall has worked for about a year.
Thanks to Jean Delvare <jdelvare@suse.de> for the note.
References:
commit 14005a371e5c14289e96a4927ffd1a827d3c9d85
commit psmisc/psmisc@1188315cd0
Signed-off-by: Craig Small <csmall@dropbear.xyz>
This patch is an outgrowth of that commit shown below.
Many additional potential segmentation faults might be
encountered if interactive commands are opened up to a
user when a '-p' switch has a single non-existent pid.
[ always the 'k', 'L', 'r', 'Y' keys & maybe 'v' too ]
So, this patch will restrict such a loser (oops, user)
to a reduced subset of normal commands until he/she/it
quits then restarts top with something to be displayed
or issues the '=' command overriding that '-p' switch.
Reference(s):
commit f57a0301e3adfa5fd456404a200182c7f21da03a
Signed-off-by: Jim Warner <james.warner@comcast.net>
This patch is an outgrowth of that commit shown below.
Many additional potential segmentation faults might be
encountered if interactive commands are opened up to a
user when a '-p' switch has a single non-existent pid.
[ always the 'k', 'L', 'r', 'Y' keys & maybe 'v' too ]
So, this patch will restrict such a loser (oops, user)
to a reduced subset of normal commands until he/she/it
quits then restarts top with something to be displayed
or issues the '=' command overriding that '-p' switch.
Reference(s):
commit d3203d99dd7f4508447ea3b4b056a9d134229d10
Signed-off-by: Jim Warner <james.warner@comcast.net>
This patch fixes a nearly decade old bug discovered by
Frederik Deweerdt. His merge request shown below would
be an adequate solution except for iterative overhead.
This alternate patch will represent substantially less
overhead for an admittedly extremely rare possibility.
Reference(s):
https://gitlab.com/procps-ng/procps/-/merge_requests/114
And-thanks-to: Frederik Deweerdt <fdeweerdt@fastly.com>
Signed-off-by: Jim Warner <james.warner@comcast.net>
