リアルタイムプロセス管理

この記事では起動時だけではなく、リアルタイムにプロセススレッドの優先度を変更する情報を記載しています。個別のプロセスあるいは特定のグループによって実行された全てのプロセスの CPU やメモリなどのリソースの使用量を制御する方法を説明します。

新しいプロセッサの多くは大量の動画や音声を同時に再生できるほどの能力を持っていますが、それでもスレッドがプロセッサを占有して他の処理に遅延が発生する可能性が全く無いわけではありません。そのような場合、音声や動画にずれが生じてしまう恐れがあり、単に趣味で音楽を聞いているだけなら煩わしいだけですみますが、作曲したり動画を編集して生業を立てている人にとっては非常に深刻な問題です。

一番簡単な解決方法は音声や動画のプロセスに高い優先度を割り当てることです。しかしながら、通常ユーザーは比較的高い nice 値 (つまり優先度は低い) しか設定できず、0未満の低い nice 値を設定してプロセスを起動できるのは root だけとなっています。これによって通常ユーザーはシステムの重要な役割を担っているプロセスの優先度を下げられないようになっています。この制限は特にユーザーが複数存在するマシンで重要です。

設定

デフォルトで、Arch ではリアルタイムの優先度変更が有効になっており、ユーザーから簡単に設定を編集することができます。例えば、0以下の nice 値をユーザーが設定できるようにするには、PAM によって設定されているデフォルトのハードリミットを調整する必要があります。

pam

公式リポジトリの pam パッケージには Linux カーネルの pluggable authentication modules が入っています。

ノート: カスタムカーネルを使っている場合、"preemptible kernel" の設定が有効になっていることを確認してください。標準の Arch カーネルでは設定は不要です。

PAM の設定

/etc/security/limits.conf ファイルにはシステムリソースの制限を設定する PAM モジュール pam_limits の設定が書き込まれています。このファイルを編集することで全てのプロセスや個別のグループのデフォルトの nice レベルを定義したり、メモリアドレス領域の制限をかけたりすることができます。

ノート: Systemd サービスによって初期化されたプロセスは limits.conf を無視します。.service ファイルで設定する必要があります。詳しくは man systemd.exec を見てください。

pam_limits によるリソースの制限には2つのタイプがあります: ハードリミットとソフトリミットです。ハードリミットは root によって設定されカーネルによって執行されますが、ソフトリミットについてはハードリミットが許可する範囲内でユーザーが設定できます。デフォルトでは、Arch は - リミットを使っており、ハードリミットとソフトリミットの両方が参照されます。

The default Arch Linux settings set the maximum real-time priority allowed for non-priveleged processes to 0, the maximum nice priority allowed to raise to 0, and some custom settings for the audio group. Finally, the memlock item sets the maximum locked-in memory address space to 40,000 KiB. These defaults are shown below:

*               -       rtprio          0
*               -       nice            0
@audio          -       rtprio          65
@audio          -       nice           -10
@audio          -       memlock         40000

An example for why one might want to alter these settings is to get high-performance audio working. The defaults are permissive enough to get jack-server running with hydrogen or ardour. However, for higher performance audio applications it might be necessary to redefine the values for rt_prio from 65 to 80 or even higher! The following settings work well with ardour:

@audio          -       rtprio          70
@audio          -       memlock         250000

詳しくはプロオーディオを見てください。

PAM limits で設定できる値は星の数ほどあります。ここで説明しているのはあくまで概要だけなので、より深く理解するために man 5 limits.conf のページを読むことを強く推奨します。

ハードリアルタイムとソフトリアルタイム

Realtime is a synonym for a process which has the capability to run in time without being interrupted by any other process. However, cycles can occasionally be dropped despite this. Low power supply or a process with higher priority could be a potential cause. To solve this problem, there is a scaling of realtime quality. This article deals with soft realtime. Hard realtime is usually not so much desired as it is needed. An example could be made for car's ABS (anti-lock braking system). This can not be "rendered" and there is no second chance.

Power is nothing without control

The realtime-lsm module granted the right to get higher capabilities to users belonging to a certain UID. The rlimit way works similar, but it can be controlled graduated finer. There is a new functionality in PAM which can be used to control the capabilities on a per user or a per group level. In the current version (0.80-2) these values are not set correctly out of the box and still create problems. With PAM you can grant realtime priority to a certain user or to a certain user group. PAM's concept makes it imaginable that there will be ways in the future to grant rights on a per application level; however, this is not yet possible.

Tips and tricks

PAM-enabled login

参照: ログイン時に X を起動。

For your system to use PAM limits settings you have to use a pam-enabled login method/manager. Nearly all graphical login managers are pam-enabled, and it now appears that the default Arch login is pam-enabled as well. You can confirm this by searching /etc/pam.d:

$ grep pam_limits.so /etc/pam.d/*

If you get nothing, you are whacked. But you will, as long as you have a login manager (and now PolicyKit). We want an output like this one:

/etc/pam.d/crond:session   required    pam_limits.so
/etc/pam.d/login:session		required	pam_limits.so
/etc/pam.d/polkit-1:session         required        pam_limits.so
/etc/pam.d/system-auth:session   required  pam_limits.so
/etc/pam.d/system-services:session   required    pam_limits.so

So we see that login, PolicyKit, and the others all require the pam_limits.so module. This is a good thing, and means PAM limits will be enforced.

コンソール/自動ログイン

参照: 仮想端末に自動ログイン。

グラフィカルログインをしない場合でも、方法はあります。(coreutils に含まれている) su の pam を編集してください [1]:

/etc/pam.d/su

 ...
 session              required        pam_limits.so

参考

RLIMIT 定義

RLIMIT_AS: The maximum size of the process’s virtual memory (address space) in bytes. This limit affects calls to brk(2), mmap(2) and mremap(2), which fail with the error ENOMEM upon exceeding this limit. Also automatic stack expansion will fail (and generate a SIGSEGV that kills the process if no alternate stack has been made available via sigaltstack(2)). Since the value is a long, on machines with a 32-bit long either this limit is at most 2 GiB, or this resource is unlimited.
RLIMIT_CORE: Maximum size of core file. When 0 no core dump files are created. When non-zero, larger dumps are truncated to this size.
RLIMIT_CPU: CPU time limit in seconds. When the process reaches the soft limit, it is sent a SIGXCPU signal. The default action for this signal is to terminate the process. However, the signal can be caught, and the handler can return control to the main program. If the process continues to consume CPU time, it will be sent SIGXCPU once per second until the hard limit is reached, at which time it is sent SIGKILL. (This latter point describes Linux 2.2 through 2.6 behavior. Implementations vary in how they treat processes which continue to consume CPU time after reaching the soft limit. Portable applications that need to catch this signal should perform an orderly termination upon first receipt of SIGXCPU.)
RLIMIT_DATA: The maximum size of the process’s data segment (initialized data, uninitialized data, and heap). This limit affects calls to brk(2) and sbrk(2), which fail with the error ENOMEM upon encountering the soft limit of this resource.
RLIMIT_FSIZE: The maximum size of files that the process may create. Attempts to extend a file beyond this limit result in delivery of a SIGXFSZ signal. By default, this signal terminates a process, but a process can catch this signal instead, in which case the relevant system call (e.g., write(2), truncate(2)) fails with the error EFBIG.
RLIMIT_LOCKS: (Early Linux 2.4 only) A limit on the combined number of flock(2) locks and fcntl(2) leases that this process may establish.
RLIMIT_MEMLOCK: The maximum number of bytes of memory that may be locked into RAM. In effect this limit is rounded down to the nearest multiple of the system page size. This limit affects mlock(2) and mlockall(2) and the mmap(2) MAP_LOCKED operation. Since Linux 2.6.9 it also affects the shmctl(2) SHM_LOCK operation, where it sets a maximum on the total bytes in shared memory segments (see shmget(2)) that may be locked by the real user ID of the calling process. The shmctl(2) SHM_LOCK locks are accounted for separately from the per-process memory locks established by mlock(2), mlockall(2), and mmap(2) MAP_LOCKED; a process can lock bytes up to this limit in each of these two categories. In Linux kernels before 2.6.9, this limit controlled the amount of memory that could be locked by a privileged process. Since Linux 2.6.9, no limits are placed on the amount of memory that a privileged process may lock, and this limit instead governs the amount of memory that an unprivileged process may lock.
RLIMIT_MSGQUEUE: (Since Linux 2.6.8) Specifies the limit on the number of bytes that can be allocated for POSIX message queues for the real user ID of the calling process. This limit is enforced for mq_open(3). Each message queue that the user creates counts (until it is removed) against this limit according to the formula: bytes = attr.mq_maxmsg * sizeof(struct msg_msg *) + attr.mq_maxmsg * attr.mq_msgsize where attr is the mq_attr structure specified as the fourth argument to mq_open(3). The first addend in the formula, which includes sizeof(struct msg_msg *) (4 bytes on Linux/i386), ensures that the user cannot create an unlimited number of zero-length messages (such messages nevertheless each consume some system memory for bookkeeping overhead).
RLIMIT_NICE: (since Linux 2.6.12, but see BUGS below) Specifies a ceiling to which the process’s nice value can be raised using setpriority(2) or nice(2). The actual ceiling for the nice value is calculated as 20 – rlim_cur. (This strangeness occurs because negative numbers cannot be specified as resource limit values, since they typically have special meanings. For example, RLIM_INFINITY typically is the same as -1.)
RLIMIT_NOFILE: Specifies a value one greater than the maximum file descriptor number that can be opened by this process. Attempts (open(2), pipe(2), dup(2), etc.) to exceed this limit yield the error EMFILE. (Historically, this limit was named RLIMIT_OFILE on BSD.)
RLIMIT_NPROC: The maximum number of processes (or, more precisely on Linux, threads) that can be created for the real user ID of the calling process. Upon encountering this limit, fork(2) fails with the error EAGAIN.
RLIMIT_RSS: Specifies the limit (in pages) of the process’s resident set (the number of virtual pages resident in RAM). This limit only has effect in Linux 2.4.x, x < 30, and there only affects calls to madvise(2) specifying MADV_WILLNEED.
RLIMIT_RTPRIO: (Since Linux 2.6.12, but see BUGS) Specifies a ceiling on the real-time priority that may be set for this process using sched_setscheduler(2) and sched_setparam(2).
RLIMIT_RTTIME: (Since Linux 2.6.25) Specifies a limit on the amount of CPU time that a process scheduled under a real-time scheduling policy may consume without making a blocking system call. For the purpose of this limit, each time a process makes a blocking system call, the count of its consumed CPU time is reset to zero. The CPU time count is not reset if the process continues trying to use the CPU but is preempted, its time slice expires, or it calls sched_yield(2). Upon reaching the soft limit, the process is sent a SIGXCPU signal. If the process catches or ignores this signal and continues consuming CPU time, then SIGXCPU will be generated once each second until the hard limit is reached, at which point the process is sent a SIGKILL signal. The intended use of this limit is to stop a runaway real-time process from locking up the system.
RLIMIT_SIGPENDING: (Since Linux 2.6.8) Specifies the limit on the number of signals that may be queued for the real user ID of the calling process. Both standard and real-time signals are counted for the purpose of checking this limit. However, the limit is only enforced for sigqueue(2); it is always possible to use kill(2) to queue one instance of any of the signals that are not already queued to the process.
RLIMIT_STACK: The maximum size of the process stack, in bytes. Upon reaching this limit, a SIGSEGV signal is generated. To handle this signal, a process must employ an alternate signal stack (sigaltstack(2)).

スケジューリングポリシー

CFS は3つのスケジューリングポリシーを実装しています:

SCHED_NORMAL (別名 SCHED_OTHER): 通常のタスクに使われるスケジューリングポリシー。
SCHED_BATCH: Does not preempt nearly as often as regular tasks would, thereby allowing tasks to run longer and make better use of caches but at the cost of interactivity. This is well suited for batch jobs.
SCHED_IDLE: This is even weaker than nice 19, but its not a true idle timer scheduler in order to avoid to get into priority inversion problems which would deadlock the machine.

スケジューリングクラス

IOPRIO_CLASS_RT: リアルタイム io クラスです。RT スケジューリングクラスは、システムで何が動作しているのかに関わらず、ディスクに対する第一級のアクセス権限を与えられます。そのため RT クラスは注意して使用しないと、他のプロセスに多大な影響を与える可能性があります。ベストエフォートクラスと同じように、スケジューリングウィンドウでどれだけタイムスライスをプロセスに与えるか定義する8つのプライオリティレベルが存在します。RT スケジューリングクラスはシステム内の他のどんなクラスよりも高い優先度を持っているため、RT スケジューリングクラスのプロセスはいかなる時でも真っ先にディスクにアクセスすることができます。Thus it needs to be used with some care, one io RT process can starve the entire system. Within the RT class, there are 8 levels of class data that determine exactly how much time this process needs the disk for on each service. In the future this might change to be more directly mappable to performance, by passing in a wanted data rate instead.
IOPRIO_CLASS_BE: This is the best-effort scheduling class, which is the default for any process that hasn’t set a specific io priority. This is the default scheduling class for any process that hasn’t asked for a specific io priority. Programs inherit the CPU nice setting for io priorities. This class takes a priority argument from 0-7, with lower number being higher priority. Programs running at the same best effort priority are served in a round-robin fashion. The class data determines how much io bandwidth the process will get, it’s directly mappable to the cpu nice levels just more coarsely implemented. 0 is the highest BE prio level, 7 is the lowest. The mapping between cpu nice level and io nice level is determined as: io_nice = (cpu_nice + 20) / 5.
IOPRIO_CLASS_IDLE: This is the idle scheduling class, processes running at this level only get io time when no one else needs the disk. A program running with idle io priority will only get disk time when no other program has asked for disk io for a defined grace period. The impact of idle io processes on normal system activity should be zero. This scheduling class does not take a priority argument. The idle class has no class data, since it doesn’t really apply here.

参照