# apps.plugin
`apps.plugin` breaks down system resource usage to **processes**, **users** and **user groups**.
To achieve this task, it iterates through the whole process tree, collecting resource usage information
for every process found running.
Since Netdata needs to present this information in charts and track them through time,
instead of presenting a `top` like list, `apps.plugin` uses a pre-defined list of **process groups**
to which it assigns all running processes. This list is customizable via `apps_groups.conf`, and Netdata
ships with a good default for most cases (to edit it on your system run `/etc/netdata/edit-config apps_groups.conf`).
So, `apps.plugin` builds a process tree (much like `ps fax` does in Linux), and groups
processes together (evaluating both child and parent processes) so that the result is always a list with
a predefined set of members (of course, only process groups found running are reported).
> If you find that `apps.plugin` categorizes standard applications as `other`, we would be
> glad to accept pull requests improving the defaults shipped with Netdata in `apps_groups.conf`.
Unlike traditional process monitoring tools (like `top`), `apps.plugin` is able to account the resource
utilization of exit processes. Their utilization is accounted at their currently running parents.
So, `apps.plugin` is perfectly able to measure the resources used by shell scripts and other processes
that fork/spawn other short lived processes hundreds of times per second.
## Charts
`apps.plugin` provides charts for 3 sections:
1. Per application charts as **Applications** at Netdata dashboards
2. Per user charts as **Users** at Netdata dashboards
3. Per user group charts as **User Groups** at Netdata dashboards
Each of these sections provides the same number of charts:
- CPU utilization (`apps.cpu`)
- Total CPU usage
- User/system CPU usage (`apps.cpu_user`/`apps.cpu_system`)
- Disk I/O
- Physical reads/writes (`apps.preads`/`apps.pwrites`)
- Logical reads/writes (`apps.lreads`/`apps.lwrites`)
- Open unique files (if a file is found open multiple times, it is counted just once, `apps.files`)
- Memory
- Real Memory Used (non-shared, `apps.mem`)
- Virtual Memory Allocated (`apps.vmem`)
- Minor page faults (i.e. memory activity, `apps.minor_faults`)
- Processes
- Threads running (`apps.threads`)
- Processes running (`apps.processes`)
- Carried over uptime (since the last Netdata Agent restart, `apps.uptime`)
- Minimum uptime (`apps.uptime_min`)
- Average uptime (`apps.uptime_average`)
- Maximum uptime (`apps.uptime_max`)
- Pipes open (`apps.pipes`)
- Swap memory
- Swap memory used (`apps.swap`)
- Major page faults (i.e. swap activity, `apps.major_faults`)
- Network
- Sockets open (`apps.sockets`)
In addition, if the [eBPF collector](/collectors/ebpf.plugin/README.md) is running, your dashboard will also show an
additional [list of charts](/collectors/ebpf.plugin/README.md#integration-with-appsplugin) using low-level Linux
metrics.
The above are reported:
- For **Applications** per target configured.
- For **Users** per username or UID (when the username is not available).
- For **User Groups** per groupname or GID (when groupname is not available).
## Performance
`apps.plugin` is a complex piece of software and has a lot of work to do
We are proud that `apps.plugin` is a lot faster compared to any other similar tool,
while collecting a lot more information for the processes, however the fact is that
this plugin requires more CPU resources than the `netdata` daemon itself.
Under Linux, for each process running, `apps.plugin` reads several `/proc` files
per process. Doing this work per-second, especially on hosts with several thousands
of processes, may increase the CPU resources consumed by the plugin.
In such cases, you many need to lower its data collection frequency.
To do this, edit `/etc/netdata/netdata.conf` and find this section:
```
[plugin:apps]
# update every = 1
# command options =
```
Uncomment the line `update every` and set it to a higher number. If you just set it to `2`,
its CPU resources will be cut in half, and data collection will be once every 2 seconds.
## Configuration
The configuration file is `/etc/netdata/apps_groups.conf`. To edit it on your system, run `/etc/netdata/edit-config apps_groups.conf`.
The configuration file works accepts multiple lines, each having this format:
```txt
group: process1 process2 ...
```
Each group can be given multiple times, to add more processes to it.
For the **Applications** section, only groups configured in this file are reported.
All other processes will be reported as `other`.
For each process given, its whole process tree will be grouped, not just the process matched.
The plugin will include both parents and children. If including the parents into the group is
undesirable, the line `other: *` should be appended to the `apps_groups.conf`.
The process names are the ones returned by:
- `ps -e` or `cat /proc/PID/stat`
- in case of substring mode (see below): `/proc/PID/cmdline`
To add process names with spaces, enclose them in quotes (single or double)
example: `'Plex Media Serv'` or `"my other process"`.
You can add an asterisk `*` at the beginning and/or the end of a process:
- `*name` _suffix_ mode: will search for processes ending with `name` (at `/proc/PID/stat`)
- `name*` _prefix_ mode: will search for processes beginning with `name` (at `/proc/PID/stat`)
- `*name*` _substring_ mode: will search for `name` in the whole command line (at `/proc/PID/cmdline`)
If you enter even just one _name_ (substring), `apps.plugin` will process
`/proc/PID/cmdline` for all processes (of course only once per process: when they are first seen).
To add processes with single quotes, enclose them in double quotes: `"process with this ' single quote"`
To add processes with double quotes, enclose them in single quotes: `'process with this " double quote'`
If a group or process name starts with a `-`, the dimension will be hidden from the chart (cpu chart only).
If a process starts with a `+`, debugging will be enabled for it (debugging produces a lot of output - do not enable it in production systems).
You can add any number of groups. Only the ones found running will affect the charts generated.
However, producing charts with hundreds of dimensions may slow down your web browser.
The order of the entries in this list is important: the first that matches a process is used, so put important
ones at the top. Processes not matched by any row, will inherit it from their parents or children.
The order also controls the order of the dimensions on the generated charts (although applications started
after apps.plugin is started, will be appended to the existing list of dimensions the `netdata` daemon maintains).
There are a few command line options you can pass to `apps.plugin`. The list of available options can be acquired with the `--help` flag. The options can be set in the `netdata.conf` file. For example, to disable user and user group charts you should set
```
[plugin:apps]
command options = without-users without-groups
```
### Integration with eBPF
If you don't see charts under the **eBPF syscall** or **eBPF net** sections, you should edit your
[`ebpf.d.conf`](/collectors/ebpf.plugin/README.md#configure-the-ebpf-collector) file to ensure the eBPF program is enabled.
Also see our [guide on troubleshooting apps with eBPF
metrics](/docs/guides/troubleshoot/monitor-debug-applications-ebpf.md) for ideas on how to interpret these charts in a
few scenarios.
## Permissions
`apps.plugin` requires additional privileges to collect all the information it needs.
The problem is described in issue #157.
When Netdata is installed, `apps.plugin` is given the capabilities `cap_dac_read_search,cap_sys_ptrace+ep`.
If this fails (i.e. `setcap` fails), `apps.plugin` is setuid to `root`.
### linux capabilities in containers
There are a few cases, like `docker` and `virtuozzo` containers, where `setcap` succeeds, but the capabilities
are silently ignored (in `lxc` containers `setcap` fails).
In these cases ()`setcap` succeeds but capabilities do not work), you will have to setuid
to root `apps.plugin` by running these commands:
```sh
chown root:netdata /usr/libexec/netdata/plugins.d/apps.plugin
chmod 4750 /usr/libexec/netdata/plugins.d/apps.plugin
```
You will have to run these, every time you update Netdata.
## Security
`apps.plugin` performs a hard-coded function of building the process tree in memory,
iterating forever, collecting metrics for each running process and sending them to Netdata.
This is a one-way communication, from `apps.plugin` to Netdata.
So, since `apps.plugin` cannot be instructed by Netdata for the actions it performs,
we think it is pretty safe to allow it have these increased privileges.
Keep in mind that `apps.plugin` will still run without escalated permissions,
but it will not be able to collect all the information.
## Application Badges
You can create badges that you can embed anywhere you like, with URLs like this:
```
https://your.netdata.ip:19999/api/v1/badge.svg?chart=apps.processes&dimensions=myapp&value_color=green%3E0%7Cred
```
The color expression unescaped is this: `value_color=green>0|red`.
Here is an example for the process group `sql` at `https://registry.my-netdata.io`:
![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.processes&dimensions=sql&value_color=green%3E0%7Cred)
Netdata is able give you a lot more badges for your app.
Examples below for process group `sql`:
- CPU usage: ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.cpu&dimensions=sql&value_color=green=0%7Corange%3C50%7Cred)
- Disk Physical Reads ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.preads&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Disk Physical Writes ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.pwrites&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Disk Logical Reads ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.lreads&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Disk Logical Writes ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.lwrites&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Open Files ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.files&dimensions=sql&value_color=green%3E30%7Cred)
- Real Memory ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.mem&dimensions=sql&value_color=green%3C100%7Corange%3C200%7Cred)
- Virtual Memory ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.vmem&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Swap Memory ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.swap&dimensions=sql&value_color=green=0%7Cred)
- Minor Page Faults ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.minor_faults&dimensions=sql&value_color=green%3C100%7Corange%3C1000%7Cred)
- Processes ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.processes&dimensions=sql&value_color=green%3E0%7Cred)
- Threads ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.threads&dimensions=sql&value_color=green%3E=28%7Cred)
- Major Faults (swap activity) ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.major_faults&dimensions=sql&value_color=green=0%7Cred)
- Open Pipes ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.pipes&dimensions=sql&value_color=green=0%7Cred)
- Open Sockets ![image](https://registry.my-netdata.io/api/v1/badge.svg?chart=apps.sockets&dimensions=sql&value_color=green%3E=3%7Cred)
For more information about badges check [Generating Badges](/web/api/badges/README.md)
## Comparison with console tools
SSH to a server running Netdata and execute this:
```sh
while true; do ls -l /var/run >/dev/null; done
```
In most systems `/var/run` is a `tmpfs` device, so there is nothing that can stop this command
from consuming entirely one of the CPU cores of the machine.
As we will see below, **none** of the console performance monitoring tools can report that this
command is using 100% CPU. They do report of course that the CPU is busy, but **they fail to
identify the process that consumes so much CPU**.
Here is what common Linux console monitoring tools report:
### top
`top` reports that `bash` is using just 14%.
If you check the total system CPU utilization, it says there is no idle CPU at all, but `top`
fails to provide a breakdown of the CPU consumption in the system. The sum of the CPU utilization
of all processes reported by `top`, is 15.6%.
```
top - 18:46:28 up 3 days, 20:14, 2 users, load average: 0.22, 0.05, 0.02
Tasks: 76 total, 2 running, 74 sleeping, 0 stopped, 0 zombie
%Cpu(s): 32.8 us, 65.6 sy, 0.0 ni, 0.0 id, 0.0 wa, 1.3 hi, 0.3 si, 0.0 st
KiB Mem : 1016576 total, 244112 free, 52012 used, 720452 buff/cache
KiB Swap: 0 total, 0 free, 0 used. 753712 avail Mem
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
12789 root 20 0 14980 4180 3020 S 14.0 0.4 0:02.82 bash
9 root 20 0 0 0 0 S 1.0 0.0 0:22.36 rcuos/0
642 netdata 20 0 132024 20112 2660 S 0.3 2.0 14:26.29 netdata
12522 netdata 20 0 9508 2476 1828 S 0.3 0.2 0:02.26 apps.plugin
1 root 20 0 67196 10216 7500 S 0.0 1.0 0:04.83 systemd
2 root 20 0 0 0 0 S 0.0 0.0 0:00.00 kthreadd
```
### htop
Exactly like `top`, `htop` is providing an incomplete breakdown of the system CPU utilization.
```
CPU[||||||||||||||||||||||||100.0%] Tasks: 27, 11 thr; 2 running
Mem[||||||||||||||||||||85.4M/993M] Load average: 1.16 0.88 0.90
Swp[ 0K/0K] Uptime: 3 days, 21:37:03
PID USER PRI NI VIRT RES SHR S CPU% MEM% TIME+ Command
12789 root 20 0 15104 4484 3208 S 14.0 0.4 10:57.15 -bash
7024 netdata 20 0 9544 2480 1744 S 0.7 0.2 0:00.88 /usr/libexec/netd
7009 netdata 20 0 138M 21016 2712 S 0.7 2.1 0:00.89 /usr/sbin/netdata
7012 netdata 20 0 138M 21016 2712 S 0.0 2.1 0:00.31 /usr/sbin/netdata
563 root 20 0 308M 202M 202M S 0.0 20.4 1:00.81 /usr/lib/systemd/
7019 netdata 20 0 138M 21016 2712 S 0.0 2.1 0:00.14 /usr/sbin/netdata
```
### atop
`atop` also fails to break down CPU usage.
```
ATOP - localhost 2016/12/10 20:11:27 ----------- 10s elapsed
PRC | sys 1.13s | user 0.43s | #proc 75 | #zombie 0 | #exit 5383 |
CPU | sys 67% | user 31% | irq 2% | idle 0% | wait 0% |
CPL | avg1 1.34 | avg5 1.05 | avg15 0.96 | csw 51346 | intr 10508 |
MEM | tot 992.8M | free 211.5M | cache 470.0M | buff 87.2M | slab 164.7M |
SWP | tot 0.0M | free 0.0M | | vmcom 207.6M | vmlim 496.4M |
DSK | vda | busy 0% | read 0 | write 4 | avio 1.50 ms |
NET | transport | tcpi 16 | tcpo 15 | udpi 0 | udpo 0 |
NET | network | ipi 16 | ipo 15 | ipfrw 0 | deliv 16 |
NET | eth0 ---- | pcki 16 | pcko 15 | si 1 Kbps | so 4 Kbps |
PID SYSCPU USRCPU VGROW RGROW RDDSK WRDSK ST EXC S CPU CMD 1/600
12789 0.98s 0.40s 0K 0K 0K 336K -- - S 14% bash
9 0.08s 0.00s 0K 0K 0K 0K -- - S 1% rcuos/0
7024 0.03s 0.00s 0K 0K 0K 0K -- - S 0% apps.plugin
7009 0.01s 0.01s 0K 0K 0K 4K -- - S 0% netdata
```
### glances
And the same is true for `glances`. The system runs at 100%, but `glances` reports only 17%
per process utilization.
Note also, that being a `python` program, `glances` uses 1.6% CPU while it runs.
```
localhost Uptime: 3 days, 21:42:00
CPU [100.0%] CPU 100.0% MEM 23.7% SWAP 0.0% LOAD 1-core
MEM [ 23.7%] user: 30.9% total: 993M total: 0 1 min: 1.18
SWAP [ 0.0%] system: 67.8% used: 236M used: 0 5 min: 1.08
idle: 0.0% free: 757M free: 0 15 min: 1.00
NETWORK Rx/s Tx/s TASKS 75 (90 thr), 1 run, 74 slp, 0 oth
eth0 168b 2Kb
eth1 0b 0b CPU% MEM% PID USER NI S Command
lo 0b 0b 13.5 0.4 12789 root 0 S -bash
1.6 2.2 7025 root 0 R /usr/bin/python /u
DISK I/O R/s W/s 1.0 0.0 9 root 0 S rcuos/0
vda1 0 4K 0.3 0.2 7024 netdata 0 S /usr/libexec/netda
0.3 0.0 7 root 0 S rcu_sched
FILE SYS Used Total 0.3 2.1 7009 netdata 0 S /usr/sbin/netdata
/ (vda1) 1.56G 29.5G 0.0 0.0 17 root 0 S oom_reaper
```
### why does this happen?
All the console tools report usage based on the processes found running *at the moment they
examine the process tree*. So, they see just one `ls` command, which is actually very quick
with minor CPU utilization. But the shell, is spawning hundreds of them, one after another
(much like shell scripts do).
### What does Netdata report?
The total CPU utilization of the system:
![image](https://cloud.githubusercontent.com/assets/2662304/21076212/9198e5a6-bf2e-11e6-9bc0-6bdea25befb2.png)
***Figure 1**: The system overview section at Netdata, just a few seconds after the command was run*
And at the applications `apps.plugin` breaks down CPU usage per application:
![image](https://cloud.githubusercontent.com/assets/2662304/21076220/c9687848-bf2e-11e6-8d81-348592c5aca2.png)
***Figure 2**: The Applications section at Netdata, just a few seconds after the command was run*
So, the `ssh` session is using 95% CPU time.
Why `ssh`?
`apps.plugin` groups all processes based on its configuration file.
The default configuration has nothing for `bash`, but it has for `sshd`, so Netdata accumulates
all ssh sessions to a dimension on the charts, called `ssh`. This includes all the processes in
the process tree of `sshd`, **including the exited children**.
> Distributions based on `systemd`, provide another way to get cpu utilization per user session
> or service running: control groups, or cgroups, commonly used as part of containers
> `apps.plugin` does not use these mechanisms. The process grouping made by `apps.plugin` works
> on any Linux, `systemd` based or not.
#### a more technical description of how Netdata works
Netdata reads `/proc//stat` for all processes, once per second and extracts `utime` and
`stime` (user and system cpu utilization), much like all the console tools do.
But it also extracts `cutime` and `cstime` that account the user and system time of the exit children of each process.
By keeping a map in memory of the whole process tree, it is capable of assigning the right time to every process, taking
into account all its exited children.
It is tricky, since a process may be running for 1 hour and once it exits, its parent should not
receive the whole 1 hour of cpu time in just 1 second - you have to subtract the cpu time that has
been reported for it prior to this iteration.
It is even trickier, because walking through the entire process tree takes some time itself. So,
if you sum the CPU utilization of all processes, you might have more CPU time than the reported
total cpu time of the system. Netdata solves this, by adapting the per process cpu utilization to
the total of the system. [Netdata adds charts that document this normalization](https://london.my-netdata.io/default.html#menu_netdata_submenu_apps_plugin).