2010-2022 © Atoptool.nl

Important notice before download

• By downloading the software you agree with the license-conditions of the GPL (an open source license). You are strongly advised to read these conditions before you download the software.
• The license states that the software is offered to you "as is", which means that guarantee and liability is limited.
• If you have any questions or remarks, please contact the maintainer of this software-package.
• Disclaimer: The authors deny any warranty for any data loss and damage whatsoever that may be caused directly or indirectly by using this software. This is experimental software. Use at your own risk.

Service activation after installation

After the package has been installed, be sure that the related services are activated.

To support handling of terminated processes using process accounting, enable the atopacct service:

  systemctl enable --now atopacct

Notice that this service should not be enabled when the package psacct or acct (depends on the Linux distribution) has been installed and enabled. In that case atop uses the daily process accounting file created via that package.

To support maintaining daily log files to keep track of long-term analysis information, enable the atop service and the related timer:

  systemctl enable --now atop
  systemctl enable --now atop-rotate.timer

 

Version 2.7.1

• Modifications for atopsar:
  1. Bug solution: atopsar does not show proper disk counters.
     

 

 

 

Version 2.7.0

• Modifications for atop:
  1. Do not show unsupported counters.
     Instead of showing a counter that is not supported by the current kernel version, not supported by a particular system (e.g. a VM) or that can not be read when atop runs unprivileged, suppress that counter entirely and use this column space to show another counter instead.
  2. Gather and show statistics per NUMA node.
     For systems with multiple NUMA nodes, cpu and memory statistics are shown per NUMA node. The number of lines consumed for these statistics can be limited.
  3. Improved discovery of network interfaces.
     Improved detection of speed and duplex mode and determination whether a network interface is physical or virtual. Besides, physical interfaces take precedence over virtual interfaces in case that the number of interfaces of a system exceeds the maximum supported by atop.
  4. Show discard counter per disk (apart from read and write counters).
  5. Show containers created by podman.
  6. Additional memory statistics on system level.
     Show counters for number of OOM kills, number of page migrations, number of compact stalls. Show sizes related to ksm sharing and zswap pool usage.
  7. Processes can be selected based on state now.
  8. No setuid by default.
     The setuid-bit is not set by default any more, although atop is still setuid-aware (only uses the privileges when really necessary in case that the setuid-bit is set). The start screen shows if atop runs with restricted view (unprivileged) or unrestricted view (privileged). With restricted view certain counter values are not available.
  9. Improved cooperation with psacct/acct package.
     In case that the psacct/acct package is installed and this service is enabled, atop will read the process accounting records from the logfile related to that package (/var/(log/)account/pacct), similar to previous versions. However, atop notices logfile rotations by the psacct/acct service now.
  10. Improved parseable output.
      Use the -Z flag in combination with the -P... flag to exchange spaces by underscores in the command name and command line to ease parsing the output.
  11. Various bugfixes and minor improvements.
• Modifications for atopsar:
  1. Show the number of OOM kills in the -s report.
     
• General:
  1. The format of the raw file is incompatible with previous versions. Raw files from previous versions can be converted to the new layout with the atopconvert command.

 

Version 2.6.0

• Modifications for atop:
  1. Sort threads within process.
     Individual threads within a process (key 'y') are shown in the order of their TID. With the additional key 'Y' threads within a process are sorted based on the current sort criterium.
  2. Show the WCHAN per thread.
     With the key 'W' is active, the wait channel is determined per thread and shown with the scheduling information (key 's').
  3. Show runqueue delay per thread/process.
     For every process (cumulated) and every thread the time spent waiting on the runqueue is shown as value 'RDELAY'.
  4. Show other modified/new metrics:
     PSI pressure during interval time is shown (system-level general).
     Counter 'avio' (average I/O time) for fast disks in microseconds or nanoseconds (system-level disk).
     Counter 'zfarc' showing the current size of the ARC (cache) for ZFSonlinux (system-level memory).
     Counter 'swcac' showing the current size of the swap cache (system-level swap).
     Counter LOCKSZ showing the virtual locked memory size (process-level memory).
     
  5. Bug fixes:
     Prevent race condition in cooperation with atopacctd.
     Avoid crash with very high load average values.
     Fix errors in reading InfiniBand states.
     
• Modifications for atop and atopsar:
  1. Support of rawlog files spanning multiple days.
     The time specification to define the begin time (-b flag) and end time (-e flag) of both atop and atopsar is [YYYYMMDD]hhmm now.
     This also applies to the branch key ('b') when using atop interactively.
  2. Support reading rawlog from a pipe.
     Rawlog data can be read from a pipe by atop and atopsar with the option '-r -'. In this way, concatenated rawlog data from the command atopcat can be directly processed.
• New program atopcat:
  1. The program atopcat concatenates rawlog files and writes this data to stdout. The output can be redirected to a new rawlog file or to atop/atopsar directly via a pipe (see option '-r -' of these commands). In this way e.g. weekly or monthly reports or extractions can be created.
• General:
  1. The format of the raw file is compatible with version 2.5.

 

Version 2.5.0

• Modifications for atop:
  1. Avoid using perf counters in VM
     New option 'perfevents' in the atoprc file to define if the Instructions Per Cycle (IPC) should be retrieved via the 'perf' counters. Specify value 'auto' (default), 'enable' or 'disable'. In case of 'auto', the Instructions Per Cycle counter will not be retrieved on VMs due to the overhead of reading this counter in a guest.
  2. Improve daily rotation of logfile for systemd-based systems.
     Exchange the files /etc/cron.d/atop, /etc/logrotate.d/psaccs_atop, /etc/logrotate.d/psaccu_atop, /usr/share/atop/atop.daily, /var/log/atop/dummy_after and /var/log/atop/dummy_before by atop-rotate.service and atop-rotate.timer.
  3. Bug fixes:
     Correction of PSI value 'io full'.
     Use smaps_rollup (if present) instead of smaps for better performance.
     Use writev instead of 3 single writes to avoid logfile corruption.
     Modified handling of error messages (remain visible after window is closed).
     Correct CID for containers created by Kubernetes.
     Allow reading a logfile via a named pipe (non-seekable).
     
• Modifications for atopconvert:
  1. Show version of input file (even if only one parameter specified).
• General:
  1. The format of the raw file is compatible with version 2.4.

 

Version 2.4.0

• Modifications for atop:
  1. Support for Nvidia GPU statistics:
     A separate daemon atopgpud has been developed to maintain statistics of Nvidia GPUs in the system. When this daemon is running, atop will connect to this daemon and will get/show the GPU statistics with every interval on system level and process level (key 'e').
     The atopgpud daemon will only be active when it is explicitly enabled (see man pages for more info).
  2. Support for Infiniband statistics:
     Counters showing the bandwidth utilization of Infiniband ports (system level) are added.
  3. Support for Pressure Stall Information (PSI):
     The average pressure percentages of CPU, memory and I/O are reported for the last 10, 60 and 300 seconds. These percentage are provided from kernel 4.20 onwards.
  4. Faster startup of atop:
     The allocation strategy for processes and threads has been improved, which will specifically be noticable on larger systems with lots of multi-threaded processes.
  5. Configurable options for atop running in the background:
     Default options of script atop.daily can be overruled by /etc/default/atop file.
  6. CPU Instructions Per Cycle (IPC):
     For every CPU, the average number of instructions per CPU cycle is shown as 'ipc'. Besides, a column is shown with the number of cycles executed per second as `cycl` (shown as effective Hz frequency).
  7. Various fixes:
     Various NFS counters corrected.
     Recognition of nvme and nbd disks.
     Recognition of DEADLINE scheduling policy.
     Proper handling of memory locking (improper handling caused malloc failures in previous versions) .
• New program atopconvert:
  1. This program can be used to convert the layout of older raw files to newer raw files. Raw files from atop 2.0 onwards can be converted to the newest version (now: 2.4).
• Modifications for atopsar:
  1. Reports can be printed now about the utilization of GPUs and Infiniband, and about the PSI percentages.
     
• General:
  1. The format of the raw file is incompatible with earlier versions. However, old raw files can be converted now by the program atopconvert to be read or extended by newer atop versions.

 

Version 2.3.0

• Modifications for atop:
  1. Support for Docker containers:
     When Docker containers are actively used in a system, the containerid (CID of 12 positions) is shown for every process. With the key 'J', it is possible to filter all processes with a particular CID. With the key 'j', all processes are accumulated per container so the load of every individual container on the host system can be watched.
  2. Improved gathering of process data:
     With the key 'a', you can switch between watching all processes or watching only the active processes (i.e. active during the interval). In earlier versions of atop, a switch from active processes to all processes could only be effectuated for the next interval. In this new version, switching is immediately effectuated within the current interval.
  3. Improved memory figures for processes:
     With the key 'm' (show memory figures of processes), automatically all processes are shown now instead of only the processes that were active, independent of the current choice made with key 'a'. After all, for the analysis of memory occupation it does not matter if a process was active or not active during the interval.
     Similar behavior is shown when the key 'M' is active (sorting on memory).
  4. Variable width for PID column:
     Earlier versions of atop used a fixed column width of 5 positions for the PID, which produced ugly output on systems that use larger PIDs. Now the column width is determined dynamically depending on the highest PID number allowed for the current system (this is valid for the columns PID, PPID and TID).
  5. Minor improvements:
     Better handling when started via sudo (not entering stopped state).
     Handling of disk drives starting with 'xvd...'.
• Modifications for atopacct:
  1. Various improvements:
     Better error handling of Netlink interface.
     Immediate reaction when receiving SIGTERM signal.
     Bypass for kernel bug 190711 (Netlink interface not available).
     Bypass for kernel bug 190271 (process accounting does not always start).
     
• Modifications for atopsar:
  1. Various bug solutions:
     Top-3 of resource consuming processes only shows active processes now.
     
• General:
  1. The format of the raw file is upward compatible with version 2.2-3.

 

Version 2.2.6

• Modifications for atop:
  1. Various improvements:
     Better handling when started via sudo (not entering stopped state).
     Handling of disk drives starting with 'xvd...'.
• Modifications for atopacct:
  1. Various improvements:
     Better error handling of Netlink interface.
     Immediate reaction when receiving SIGTERM signal.
     Bypass for kernel bug 190711 (Netlink interface not available).
     Bypass for kernel bug 190271 (process accounting does not always start).
     
• Modifications for atopsar:
  1. Various bug solutions:
     Top-3 of resource consuming processes should only show active processes.
     
• General:
  1. The format of the raw file is compatible with version 2.2-3.

 

Version 2.2

• Modifications for atop:
  1. New keys for process selection:
     '/' to select processes containing a particular search string in the command line arguments.
     'I' to select processes running with particular PIDs.
  2. New counters for NFS activity:
     On system level, counters are maintained about NFS traffic. Three new labels are introduced for this purpose:
     'NFS' for NFS server statistics (overall).
     'NFC' for NFS client statistics (overall).
     'NFM' for NFS client statistics per mounted NFS filesystem (so multiple lines possible).
  3. New counters for OpenVZ containers:
     CPU consumption and memory occupation are shown per container on system level (label 'CON').
     Container ID and virtual PID are shown on process level (when using key 'v' or 's').
  4. Bug fixes:
     The maximum transfer rate larger than 1 Gbit for ethernet interfaces is now correctly shown.
     The maximum transfer rate for wifi connections is now shown.
     For interfaces in general, the maximum transfer rate is shown as a separate column ('sp').
      
     Counters about virtual disk type 'xvd[a-z]' are shown.
      
     Several minor bugs have been solved.
• Modifications for atopacct:
  1. Avoid that process accounting is reactivated too frequently (and unnecessary).
• Modifications for atopsar:
  1. Add flags for NFS traffic ('-n' for NFS mounts, '-j' for overall NFS client activity and '-J' for overall NFS server activity).
• General:
  1. Support for systemd-based distributions has been added.
  2. The format of the raw file is incompatible with earlier versions. However, the earlier version of atop is automatically activated when accessing an older raw file.

 

Version 2.1

• Modifications for atop:
  1. New daemon atopacctd:
     This daemon has been added to control the process accounting mechanism. It takes care of truncating the original process accounting file regularly and creates small-sized shadow files to be read by the atop processes. In this way, excessive filesystem space consumption is avoided.
  2. New keys/flags:
     'G' to suppress showing/accumulating exited processes in the output and 'R' to calculate the proportional set size (PSS) of processes (optional because this is rather time-consuming).
  3. New fields:
     On system level, in the MEM line some counters are added for the space reserved for huge pages (total and in use) and the space reserved by the VMWare balloon driver (to be reclaimed by the host machine).
     On process level, the proportional set size (PSIZE) is added to the memory details and the virtual environment identification (ENVID in case of OpenVZ) is added to the scheduling/variable details.
  4. Bug fixes:
     Several bugs have been solved that caused segmentation faults (especially in combination with thread view) and floating point exceptions.
  5. Configurable colors:
     In the atoprc file, the colors can be redefined for information messages (default green), threads (default yellow), almost critical resources (default cyan) and critical resources (default red).
  6. Various minor modifications on counters, installation files and documentation (man pages).
• Modifications for atopsar:
  1. Several bugs have been solved that caused segmentation faults.
• General: The format of the raw file is incompatible with earlier versions. However, the earlier version of atop is automatically activated when accessing an older raw file.

 

Version 2.0.2

• Modifications for atop:
  1. Bug-solution: The per-process disk statistics are only shown for own processes, so not for processes running under another identity.
  2. Bug-solution: The space reserved for the RNET/SNET counters in the generic screen is too narrow, especially when requesting per-second values (key '1').
  3. Bug-solution: When per-user/per-program (key 'u'/'p') accumulation is active, no proper sorting on the network activity is done.

 

Version 2.0

• Modifications for atop:
  1. Support for netatop kernel module and netatopd daemon.
     You can download and install this kernel module and daemon as a separate and optional tar-archive (atop can perfectly run without). The module and daemon gather network utilization counters per process/thread, as described here.
  2. Support for (atop related) kernel patches removed.
  3. Show shared memory utilization on system level (added to MEM line).
  4. Adapt formula for coloring of MEM line (take shared memory into account).
  5. Minor bug fixes.
• Modifications for atopsar:
  1. Support for netatop kernel module.
• General: The format of the raw file is compatible with version 1.27.

 

Version 1.27-3

• Modifications for atop:
  1. Resource consumption of threads (key 'y' and flag '-y'):
     For multi-threaded processes, the individual threads (depending on key 'a', only the acive ones or all) are shown in a dififerent color with their TID and resource utilization.
  2. Improved color handling for system-level values:
     Only show the relevant values within a resource line in cyan/red and not the entire line.
  3. Improved handling of multiple system resources (disks, LVMs, interfaces, ...):
     Sorting can be suppressed to see these resources always in the same order (key 'F' and flag -F). Furthermore selections can be made of these resources by specifying a regular expression (key 'S').
  4. Improved screen handling:
     The Page-Up/Page-Down keys and Arrow-Up/Arrow-Down keys (per line) can be used to browse the process/thread list. The Arrow-Right/Arrow-Left keys can be used for horizontal scrolling to view the full command line (when key 'c' is active).
  5. Improved handling of process accounting:
     When the process accounting file appears to be larger than 200 MiB while only one atop-process is active, the process accounting file is truncated to zero to save disk space.
     The maximum number of processes that will be read from the process accounting file for one interval is about 72000, requiring 50 MiB of memory in atop. When this maximum value is exceeded, a warning is given that finished processes are partly ignored. This limitation avoids that atop itself consumes too much memory space.
  6. More information about memory consumption on process level:
     Swap space used and virtual usage of data, stack and shared libraries.
  7. Extended limits for system resources:
     Maximum number of CPUs now 2048 (was: 64), disks now 512 (was: 256), logical volumes now 1024 (was: 256), MD devices 256 (was: 128) and network interfaces now 128 (was: 32).
  8. Miscellaneous modifications:
     New disk types recognized (EMC power, virtual disks), no delay at startup to show error messages, various bug fixes.
• Modifications for atopsar:
  1. Various minor bug-fixes.
• General: The format of the raw file is incompatible with earlier versions. However, the earlier version of atop is automatically activated when accessing an older raw file.

 

Version 1.26-2

• Modifications for atop:
  1. An additional value is shown for the current frequency and the current scaling percentage of the CPU. When the kernel module 'cpufreq_stats' is active, average values are shown instead of current values.
  2. Additional system-level counters are shown for the total number of threads that are running, sleeping interruptible and sleeping uninterruptible.
  3. When the number of lines in a window is too small for the amount of system-level lines, the number of variable resources (like disks, interfaces, etc) is limited automatically to be able to continue.
  4. Signal SIGUSR2 can be sent to atop to take one more sample and terminate (e.g. used to speed up suspend/hibernate).
  5. Support disk type 'mmcblk'.
• Version 1.26-2 (available since november 18, 2010) contains two bug fixes related to segmentation faults (one caused by memcpy instead of memmove and one caused by not closing /proc/cpuinfo) and a bug fix for a filename consisting of 8 'y'-characters (indicating 8 days ago).
• General: The format of the raw file is incompatible with earlier versions. However, the earlier version of atop is automatically activated when accessing an older raw file.

 

Version 1.25

• Modifications for atop:
  1. Disk I/O statistics per logical volume (label LVM) and per MD device (label MDD). Number of devices to be shown can be limited, similar to hard disks.
  2. Support of suspend/hibernate (e.g. for laptops).
     At the moment of suspend/hibernate, atop is forced to take one extra sample whereafter it will be terminated. At the moment of resume/thaw, atop is started again (continue in the same raw file or -- in case of a new day -- in a new raw file).
  3. The layout of the output lines with system-level statistics can be redefined in the /etc/atoprc or ~/.atoprc file (see man-page of atoprc). Also the priority of every field with system-level info can be influenced.
  4. Automatic activation of an older version of atop/atopsar when using incompatible raw files. From version 1.25 onwards, atop and atopsar will be preserved as /usr/bin/atop-version and /usr/bin/atopsar-version after deinstallation. The current (newer) version will activate the older version when a raw file must be read that is created by that older version.
     You can preserve a version older than version 1.25 manually. E.g. if you currently use version 1.24, save atop/atopsar before deinstalling version 1.24:

     cp /usr/bin/atop /usr/bin/atop-1.24
     ln -s atop-1.24  /usr/bin/atopsar-1.24
             

     Notice that atop versions < 1.25 do not allow unknown keywords in the rc-file (so do not use keywords that are not supported by such version). From version 1.25 onwards, only a warning is given for an unknown keyword.
  5. Various minor improvements (multiple sorts are eliminated, proper sorting of accumulated user/program statistics, show more active keys in header line, support /dev/fio disks, show Tbytes i.s.o. Gbytes for large sizes).
  6. Various bugfixes (buffer overflow for systems with long uptime, wrong cpu number in per-cpu statistics, wrong uid/gid for exited processes).
• Modifications for atopsar:
  1. Disk I/O statistics per logical volume (flag -l) and per MD device (flag -f).
• General: The format of the raw file is incompatible with earlier versions. However, refer to modification d in the list above.

 

Version 1.24

• Modifications for atop:
  1. Complete redesign of the screen interface with support of variable number of columns depending on the window size.
     It is also possible to define your own output-line with process-level statistics by using the 'o' key.
     Typing Control-L will force a screen redraw.
  2. Introduce system-wide configuration file /etc/atoprc that can be overruled by your personal ~/.atoprc file.
  3. Improved logfile handling:
     Find the logfile of yesterday, the day before yesterday, etc. by using the flag -r [yy...].
     Branch to a specific time in the logfile by using the 'b' key.
  4. Drop the root-priviliges earlier in case of setuid-root.
  5. More process-level counters like ruid, euid, suid, fsuid and rgid, egid, sgid, fsgid. Also the date and time that a process has finished is maintained.
     More system-level counters like dirty cache size and guest CPU-utilization.
  6. Bugfixes.
• Modifications for atopsar:
  1. More system-level counters like dirty cache size and guest CPU-utilization.
  2. Find the logfile of yesterday, the day before yesterday, etc. by using the flag -r y[y...].
• General: The format of the raw file is incompatible with earlier versions.

 

Version 1.23

• Modifications for atop:
  1. Gather/show for each process the number of threads that are in the state 'running', 'interruptible sleeping' and 'non-interruptible sleeping'.
  2. Show PPID for each process.
  3. Sort network-interfaces on their busy-percentage and not on the number of transferred packages.
  4. Optionally specify the name of a specific process accounting file by using the environment variable ATOPACCT. If this variable exists but has no value, process accounting will not be used at all.
  5. Bugfixes.
• Modifications for atopsar:
  1. Option to repeat the header of a report every N lines (flag -H).
  2. Option to make summary reports by calculating the average values over a period of N samples (flag -R). For example, if the logfile contains samples of 10 minutes, the option -R6 generates a report with lines containing the averages per hour.
  3. Show the total number of threads that are in the state 'running', 'interruptible sleeping' and 'non-interruptible sleeping' (flag -P).
  4. Show the top-3 of processes that have consumed most CPU time (flag -O), most memory (flag -G), most disk accesses (flag -D) and most network accesses (flag -N only if kernel patches are installed).
  5. Show 100% usage per CPU instead of 100% for all CPUs (now consistent with the output generated by atop).
  6. Rename forks/s to clones/s (this counter reflects thread creations, not process creations).
• General: The format of the raw file is compatible with version 1.21 and 1.22.

 

Version 1.22

• Modifications for atop:
  1. Addition of colors (default) and markers (flag -M) to reports generated by the command atopsar to highlight that resources are (almost) overloaded.
     Personal default flags for atopsar can also be specified in the file ~/.atoprc now.
  2. Bugfixes (mainly for large and busy machines):

• Modifications for atopsar:
  1. Always use line-buffering on stdout for atopsar, even for pipes and files.
  2. Solve huge exit-codes on 64-bits machines.
  3. Avoid wrong matches on machines with lots of process-creations (reuse of same PID by other program).
  4. Detect process-exit while reading its administration.
  5. Several minor bugfixes.

• General: The format of the raw file is compatible with version 1.21.

 

Version 1.21

• Introduction of the integrated command atopsar to produce system activity reports similar to the UNIX command sar. These reports can be generated from the (daily) raw files that are written by atop or stand-alone by specifying an interval on the command line.
• Minor bugfixes.
• Warning:
  The format of the raw file is incompatible with previous versions because many counters are added for the command atopsar!

 

Version 1.20

• Produce unformatted output to be processed by scripts.
• Show steal-percentage for virtual machines.
• Show system-level counters for disks within virtual machines.
• Restructured scripts for daily logging.
• Adapted spec-file and Makefile for improved installation.
• Bugfix: Allow more than 65535 processes to be logged per interval.
• Minor bugfixes.
• Restructure layout of raw logfile.
  Warning: The format of the raw file is incompatible with previous versions!
• Support standard disk I/O statistics per process (kernel 2.6.20 and higher).
  For full accounting of disk I/O and network I/O per process, still the kernel patches are preferred. When these patches are not installed, the standard disk I/O statistics will be shown.