User Cgroups
This feature relies on cgroups v2
, which is a kernel option.
To set a limit to "unlimited", pass a limit with blank value. Setting a limit to be blank tells the system it should be unlimited..
Enable
See the full install guide here
Debian 11 will have it enabled in the kernel by default.
CentOS 8, Debian 10, and older, will need to have it turned on in the kernel (requires reboot)
cd /usr/local/directadmin/custombuild
./build grub_conf
./build apache
./build php n
Anything older does not support cgroups.
DirectAdmin will hide or ignore values if the system has not created: /sys/fs/cgroup/user.slice
JSON Packages
Show package:
CMD_SHOW_USER_PACKAGE?package=gold&json=yes
CMD_SHOW_RESELLER_PACKAGE?package=gold&json=yes
New array, eg:
"cgroup":
{
"options":
{
"CPUQuota":
{
"cgroup_type": "percent",
"default": "",
"desc": "Percent of CPU Core. >100% for more cores.",
"name": "CPUQuota",
"placeholder": "400%",
"string": "CPU Quota",
"type": "text"
},
"IOReadBandwidthMax":
{
"cgroup_type": "io",
"default": "",
"desc": "Max rate data can be read from a disk.",
"minimum": "256K",
"name": "IOReadBandwidthMax",
"placeholder": "5M",
"string": "IO Read Bandwidth Max",
"type": "text"
},
"IOReadIOPSMax":
{
"cgroup_type": "iops",
"default": "",
"desc": "Max disk read operations per second.",
"minimum": "256",
"name": "IOReadIOPSMax",
"placeholder": "1K",
"string": "IOPS Read Max",
"type": "text"
},
"IOWriteBandwidthMax":
{
"cgroup_type": "io",
"default": "",
"desc": "Max rate data can be written to a disk.",
"minimum": "256K",
"name": "IOWriteBandwidthMax",
"placeholder": "5M",
"string": "IO Write Bandwidth Max",
"type": "text"
},
"IOWriteIOPSMax":
{
"cgroup_type": "iops",
"default": "",
"desc": "Max disk write operations per second.",
"minimum": "256",
"name": "IOWriteIOPSMax",
"placeholder": "1K",
"string": "IOPS Write Max",
"type": "text"
},
"MemoryHigh":
{
"cgroup_type": "bytes",
"default": "",
"desc": "Throttle Memory above this limit.",
"name": "MemoryHigh",
"placeholder": "1G",
"string": "Memory High",
"type": "text"
},
"TasksMax":
{
"cgroup_type": "int",
"default": "",
"desc": "Max Tasks that may be created in the unit",
"minimum": "5",
"name": "TasksMax",
"placeholder": "4915",
"string": "Tasks Max",
"type": "text"
}
},
"saved":
{
"CPUQuota":
{
"cgroup_type": "percent",
"default": "",
"desc": "Percent of CPU Core. >100% for more cores.",
"name": "CPUQuota",
"placeholder": "400%",
"string": "CPU Quota",
"type": "text",
"value": "400%"
}
}
},
Where options
are what can be used, with info about the item.. and saved
is which of those items actually exist in the package, including the value
. Items that have a minimum
are enforced at the back-end and blocks the save entirely (Creation/Modification)
Option variables
Each group option will have items:
- cgroup_type: see below. The specific format for a given option.
- default: the default value, usually blank (unlimited)
- desc: longer description, translated in the internal gettext
- name: the form name to pass, matches the system's cgroup option name.
- placeholder: for empty inputs, an example/background text
- string: the human readable name (shorter), translated in the internal gettext
- type:text form field type (eg: future might have checkbox, for example, no others for now)
- value: Only for saved values in the package itself.
options
vs saved
does have mostly redundant info. Essentially, just the "value" will exist for the saved
array.
Package Defaults:
CMD_SHOW_USER_PACKAGE?package=unlimited
CMD_SHOW_RESELLER_PACKAGE?package=unlimited
Same as listing a specific package, but the "saved" array will not be passed, since it's just the default list.
Save Package
CMD_MANAGE_USER_PACKAGES
CMD_MANAGE_RESELLER_PACKAGES
method: POST
plus all new values.
To set an option to be "unlimited", it must be set with a blank value. Eg:
CPUQuota=
will removing any cpu limits.
As before, all values must be sent when saving a package. The requirement to save these options is not required, but saving them once, then not passing them later, will not delete them from User settings. cgroup settings are not "deleted" from the system, they must be saved to unlimited (blank value) to remove the limits.
Cgroup types
DA uses a variable called: cgroup_type
, as defined:
percent: 1% - 10000%
int: 1 - 1000000
bytes: 1, 2K, 5M, 2G, 3T
io: same as "bytes", but internally DA adds the list of devices to the value. Packages/Users only set the number, not the device, eg: IOReadBandwidthMax=5M
iops: same as "io"
Users
Create/Modify Users/Resellers use the same formatting as packages. Pass the values you'd like to set during:
- Custom User Creation (
CMD_ACCOUNT_USER
, without setting package) - User Modification (
CMD_MODIFY_USER
/CMD_MODIFY_RESELLER
) - Show User:
CMD_SHOW_USER?user=fred&json=yes
Reseller
When setting Cgroup values for a Reseller, they're saved into that Resller's reseller.conf
file. The values set here are optional. If they are blank, the Reseller is free to set (or not set) values for their Users as desired. If they are set, then the value set is the maximum any of the Reseller's Users can have. For example, if the reseller.conf has: CPUQuota=400%
This implies that any User under this Reseller must have a CPUQuota set (cannot be blank) and it must not exceed 400%. All Users under the Reseller can have at most CPUQuota=400%
. This is different than the other cumulative values, like bandwidth or disk usage, where the Users are added together. This is not like that.. the values are the max for each User.
Stats
Admin Level
Show Live Slice info:
CMD_ADMIN_LIMITS?json=yes
CMD_ADMIN_LIMITS?json=yes&user=fred
Include live "rates" (values over time) in the above output, eg:
CMD_ADMIN_LIMITS?json=yes&live_rates=yes
CMD_ADMIN_LIMITS?json=yes&live_rates=yes&user=fred
Show Logged Events Cache:
This will show the total count of each area, and the date this area was later triggered:
CMD_ADMIN_LIMITS?json=yes&show=events_cache
CMD_ADMIN_LIMITS?json=yes&show=events_cache&user=fred
where the "date" (timestamp) will be the last time an event was caught, and the value is the current highest value. If User-Slice resets to zero, this count will continue to rise.
Show Events:
CMD_ADMIN_LIMITS?json=yes&show=events&user=fred
Default values of nothing passed:
range_start=1d
range_end=0s
where these values can be passed to control which logs are returned. The range_*= values are:
- if value ends in a time unit: s,m,h,d,M,y, will be that amount of time before now.
- If there are no units, the value must be a timestamp for the desired time.
The output contains an array "events", with a list of sub-arrays. Each sub-arrays' have an index of the timestamp the event was logged. The contents of the sub-array will be which category and event variable was increased, only showing the change since the last check.
eg;
{
"events":
{
"1612253281":
{
"memory.events":
{
"high": "20"
},
"pids.events":
{
"recent": "19"
},
"processes":
{
"cmd": "php-fpm74",
"cpu": "0.0",
"memory": "4884",
"pid": "157784",
"started": "21:13:43",
"time": "00:00:00"
}
}
}
}
where there were 20 new 'high" memory events, and 19 new process limit events since the last time the dataskq checked the count. The file that stores the "last check" will be at: /usr/local/directadmin/data/admin/cgroup_cache/fred/user.slice.json
where the diff from user.slice.json
vs the /sys/fs/cgroup/user.slice/user-1007.slice
are used for the triggers.
Other events[12345] values:
"cpu.stat":
{
"nr_periods": "55",
"nr_throttled": "42",
"throttled_percentage": "76.363640"
},
and:
"io.stat":
{
"dbytes": "55",
"253:0": {
"dbytes" : "55"
}
},
where
- dbytes could also be dios for discarded IOPs.
- io.stats["dbytes"] will be a sum of all devices that trigged this check (not likely to have multiple, but supported
- "253:0" is the name of the device (the io.max is used as the list to check)
- io.stats["253:0"]['dbytes"] will be the discarded bytes for this device. Currently a "rate" does not trigger an event, only dropped bytes/ios, as the rate can be viewed at any time.
Reseller/User Levels
The calls to the information by Resellers is the same as for Admins, with the difference being the call name, and that instead of "All Users", it's only the list from their users.list
file.
CMD_RESELLER_LIMITS?json=yes
CMD_RESELLER_LIMITS?json=yes&show=events&user=fred
CMD_RESELLER_LIMITS?json=yes&show=events_cache
with &user=fred
being optional for all, except show=events
, which must always have a user passed.
Users, again, can make the same calls, except all cases are as if user=fred was passed (this is done internally)
CMD_USER_LIMITS?json=yes
CMD_USER_LIMITS?json=yes&show=events
CMD_USER_LIMITS?json=yes&show=events_cache
Events cache
For all Levels, the show=events_cache
will show the count for the listed Users from the file: /usr/local/directadmin/data/admin/cgroup_cache/fred/events.cache
which holds the count for all type events for this Users. It's deleted with the monthly reset.
By default the internal newer_than
variable is set to 1 day ago, such that only file mtime timestamps newer than 24 hours are loaded.
To include older (or only newer) event.cache
files, you can specify either: newer_than=1612736565
where exact timestamp is the limit/threshold
or a unit value to be subtracted from now: newer_than=2d
specifying all newer than 2 days ago, allowing all units. Absence of a unit is not seconds, it's the timestamp.
For example:
{
"users":
{
"fred":
{
"cpu.stat":
{
"nr_periods":
{
"new": "594",
"timestamp": "1613085301",
"timestamp_diff": "59",
"total": "4412"
},
"nr_throttled":
{
"new": "594",
"timestamp": "1613085301",
"timestamp_diff": "59",
"total": "4403"
},
"throttled_percentage": "100.000000"
},
"memory.events":
{
"high":
{
"new": "16",
"timestamp": "1613085361",
"timestamp_diff": "119",
"total": "370"
}
}
}
}
}
}
Global tokens
"HAVE_CGROUP": "yes",
"CGROUP_CPUQuota": "25%",
"CGROUP_IOReadBandwidthMax": "1M",
"CGROUP_IOReadIOPSMax": "256",
"CGROUP_IOWriteBandwidthMax": "1M",
"CGROUP_IOWriteIOPSMax": "",
"CGROUP_MemoryHigh": "128M",
"CGROUP_TasksMax": "5",
If any value is unset or blank, it'll always be set here, with the value of ""