biotop

hloeffler · Feb 7, 2016 · 6f075b9 · 6f075b9
1 parent 31cd104
commit 6f075b9
Show file tree

Hide file tree

Showing 4 changed files with 489 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -67,6 +67,7 @@ Tools:
 
 - tools/[bashreadline](tools/bashreadline.py): Print entered bash commands system wide. [Examples](tools/bashreadline_example.txt).
 - tools/[biolatency](tools/biolatency.py): Summarize block device I/O latency as a histogram. [Examples](tools/biolatency_example.txt).
+- tools/[biotop](tools/biotop.py): Top for disks: Summarize block device I/O by process. [Examples](tools/biotop_example.txt).
 - tools/[biosnoop](tools/biosnoop.py): Trace block device I/O with PID and latency. [Examples](tools/biosnoop_example.txt).
 - tools/[bitesize](tools/bitesize.py): Show per process I/O size histogram. [Examples](tools/bitesize_example.txt).
 - tools/[cachestat](tools/cachestat.py): Trace page cache hit/miss ratio. [Examples](tools/cachestat_example.txt).

diff --git a/man/man8/biotop.8 b/man/man8/biotop.8
@@ -0,0 +1,90 @@
+.TH biotop 8  "2016-02-06" "USER COMMANDS"
+.SH NAME
+biotop \- Block device (disk) I/O by process top.
+.SH SYNOPSIS
+.B biotop [\-h] [\-C] [\-r MAXROWS] [interval] [count]
+.SH DESCRIPTION
+This is top for disks. 
+
+This traces block device I/O (disk I/O), and prints a per-process summary every
+interval (by default, 1 second). The summary is sorted on the top disk
+consumers by throughput (Kbytes). The PID and process name shown are measured
+from when the I/O was first created, which usually identifies the responsible
+process.
+
+For efficiency, this uses in-kernel eBPF maps to cache process details (PID and
+comm) by I/O request, as well as a starting timestamp for calculating I/O
+latency, and the final summary.
+
+This works by tracing various kernel blk_*() functions using dynamic tracing,
+and will need updating to match any changes to these functions.
+
+Since this uses BPF, only the root user can use this tool.
+.SH REQUIREMENTS
+CONFIG_BPF and bcc.
+.SH EXAMPLES
+.TP
+Summarize block device I/O by process, 1 second screen refresh:
+#
+.B biotop
+.TP
+Don't clear the screen:
+#
+.B biotop -C
+.TP
+5 second summaries, 10 times only:
+#
+.B biotop 5 10
+.SH FIELDS
+.TP
+loadavg:
+The contents of /proc/loadavg
+.TP
+PID
+Cached process ID, if present. This usually (but isn't guaranteed) to identify
+the responsible process for the I/O.
+.TP
+COMM
+Cached process name, if present. This usually (but isn't guaranteed) to identify
+the responsible process for the I/O.
+.TP
+D
+Direction: R == read, W == write.
+.TP
+MAJ
+Major device number.
+.TP
+MIN
+Minor device number.
+.TP
+DISK
+Disk device name.
+.TP
+I/O
+Number of I/O during the interval.
+.TP
+Kbytes
+Total Kbytes for these I/O, during the interval.
+.TP
+AVGms
+Average time for the I/O (latency) from the issue to the device, to its
+completion, in milliseconds.
+.SH OVERHEAD
+Since block device I/O usually has a relatively low frequency (< 10,000/s),
+the overhead for this tool is expected to be low or negligible. For high IOPS
+storage systems, test and quantify before use.
+.SH SOURCE
+This is from bcc.
+.IP
+https://github.com/iovisor/bcc
+.PP
+Also look in the bcc distribution for a companion _examples.txt file containing
+example usage, output, and commentary for this tool.
+.SH OS
+Linux
+.SH STABILITY
+Unstable - in development.
+.SH AUTHOR
+Brendan Gregg
+.SH SEE ALSO
+biosnoop(8), biolatency(8), iostat(1)
diff --git a/tools/biotop.py b/tools/biotop.py
@@ -0,0 +1,211 @@
+#!/usr/bin/python
+# @lint-avoid-python-3-compatibility-imports
+#
+# biotop  block device (disk) I/O by process.
+#         For Linux, uses BCC, eBPF.
+#
+# USAGE: biotop.py [-h] [-C] [-r MAXROWS] [interval] [count]
+#
+# This uses in-kernel eBPF maps to cache process details (PID and comm) by I/O
+# request, as well as a starting timestamp for calculating I/O latency.
+#
+# Copyright 2016 Netflix, Inc.
+# Licensed under the Apache License, Version 2.0 (the "License")
+#
+# 06-Feb-2016   Brendan Gregg   Created this.
+
+from __future__ import print_function
+from bcc import BPF
+from time import sleep, strftime
+import argparse
+import signal
+from subprocess import call
+
+# arguments
+examples = """examples:
+    ./biotop            # block device I/O top, 1 second refresh
+    ./biotop -C         # don't clear the screen
+    ./biotop 5          # 5 second summaries
+    ./biotop 5 10       # 5 second summaries, 10 times only
+"""
+parser = argparse.ArgumentParser(
+    description="Block device (disk) I/O by process",
+    formatter_class=argparse.RawDescriptionHelpFormatter,
+    epilog=examples)
+parser.add_argument("-C", "--noclear", action="store_true",
+    help="don't clear the screen")
+parser.add_argument("-r", "--maxrows", default=20,
+    help="maximum rows to print, default 20")
+parser.add_argument("interval", nargs="?", default=1,
+    help="output interval, in seconds")
+parser.add_argument("count", nargs="?", default=99999999,
+    help="number of outputs")
+args = parser.parse_args()
+interval = int(args.interval)
+countdown = int(args.count)
+maxrows = int(args.maxrows)
+clear = not int(args.noclear)
+
+# linux stats
+loadavg = "/proc/loadavg"
+diskstats = "/proc/diskstats"
+
+# signal handler
+def signal_ignore(signal, frame):
+    print()
+
+# load BPF program
+b = BPF(text="""
+#include <uapi/linux/ptrace.h>
+#include <linux/blkdev.h>
+
+// a value of one map, and a key for another:
+struct who_t {
+    u32 pid;
+    char name[TASK_COMM_LEN];
+};
+
+struct info_t {
+    u32 pid;
+    int type;
+    int major;
+    int minor;
+    char name[TASK_COMM_LEN];
+};
+
+struct val_t {
+    u64 bytes;
+    u64 us;
+    u32 io;
+};
+
+BPF_HASH(start, struct request *);
+BPF_HASH(whobyreq, struct request *, struct who_t);
+BPF_HASH(counts, struct info_t, struct val_t);
+
+// cache PID and comm by-req
+int trace_pid_start(struct pt_regs *ctx, struct request *req)
+{
+    struct who_t who = {};
+
+    if (bpf_get_current_comm(&who.name, sizeof(who.name)) == 0) {
+        who.pid = bpf_get_current_pid_tgid();
+        whobyreq.update(&req, &who);
+    }
+
+    return 0;
+}
+
+// time block I/O
+int trace_req_start(struct pt_regs *ctx, struct request *req)
+{
+    u64 ts;
+
+    ts = bpf_ktime_get_ns();
+    start.update(&req, &ts);
+
+    return 0;
+}
+
+// output
+int trace_req_completion(struct pt_regs *ctx, struct request *req)
+{
+    u64 *tsp;
+
+    // fetch timestamp and calculate delta
+    tsp = start.lookup(&req);
+    if (tsp == 0) {
+        return 0;    // missed tracing issue
+    }
+
+    struct who_t *whop;
+    struct val_t *valp, zero = {};
+    u64 delta_us = (bpf_ktime_get_ns() - *tsp) / 1000;
+
+    // setup info_t key
+    struct info_t info = {};
+    info.major = req->rq_disk->major;
+    info.minor = req->rq_disk->first_minor;
+    info.type = req->cmd_flags & REQ_WRITE;
+    whop = whobyreq.lookup(&req);
+    if (whop == 0) {
+        // missed pid who, save stats as pid 0
+        valp = counts.lookup_or_init(&info, &zero);
+    } else {
+        info.pid = whop->pid;
+        __builtin_memcpy(&info.name, whop->name, sizeof(info.name));
+        valp = counts.lookup_or_init(&info, &zero);
+    }
+
+    // save stats
+    valp->us += delta_us;
+    valp->bytes += req->__data_len;
+    valp->io++;
+
+    start.delete(&req);
+    whobyreq.delete(&req);
+
+    return 0;
+}
+""", debug=0)
+b.attach_kprobe(event="blk_account_io_start", fn_name="trace_pid_start")
+b.attach_kprobe(event="blk_start_request", fn_name="trace_req_start")
+b.attach_kprobe(event="blk_mq_start_request", fn_name="trace_req_start")
+b.attach_kprobe(event="blk_account_io_completion",
+    fn_name="trace_req_completion")
+
+print('Tracing... Output every %d secs. Hit Ctrl-C to end' % interval)
+
+# cache disk major,minor -> diskname
+disklookup = {}
+with open(diskstats) as stats:
+    for line in stats:
+        a = line.split()
+        disklookup[a[0] + "," + a[1]] = a[2]
+
+# output
+exiting = 0
+while 1:
+    try:
+        sleep(interval)
+    except KeyboardInterrupt:
+        exiting = 1
+
+    # header
+    if clear:
+        call("clear")
+    else:
+        print()
+    with open(loadavg) as stats:
+        print("%-8s loadavg: %s" % (strftime("%H:%M:%S"), stats.read()))
+    print("%-6s %-16s %1s %-3s %-3s %-8s %5s %7s %6s" % ("PID", "COMM",
+        "D", "MAJ", "MIN", "DISK", "I/O", "Kbytes", "AVGms"))
+
+    # by-PID output
+    counts = b.get_table("counts")
+    line = 0
+    for k, v in reversed(sorted(counts.items(),
+                                key=lambda counts: counts[1].bytes)):
+
+        # lookup disk
+        disk = str(k.major) + "," + str(k.minor)
+        if disk in disklookup:
+            diskname = disklookup[disk]
+        else:
+            diskname = "?"
+
+        # print line
+        avg_ms = (float(v.us) / 1000) / v.io
+        print("%-6d %-16s %1s %-3d %-3d %-8s %5s %7s %6.2f" % (k.pid, k.name,
+            "W" if k.type else "R", k.major, k.minor, diskname, v.io,
+            v.bytes / 1024, avg_ms))
+
+        line += 1
+        if line >= maxrows:
+            break
+    counts.clear()
+
+    countdown -= 1
+    if exiting or countdown == 0:
+        print("Detaching...")
+        exit()