forked from iovisor/bcc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
zfsslower.py
executable file
·318 lines (277 loc) · 8.88 KB
/
zfsslower.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
#!/usr/bin/python
# @lint-avoid-python-3-compatibility-imports
#
# zfsslower Trace slow ZFS operations.
# For Linux, uses BCC, eBPF.
#
# USAGE: zfsslower [-h] [-j] [-p PID] [min_ms]
#
# This script traces common ZFS file operations: reads, writes, opens, and
# syncs. It measures the time spent in these operations, and prints details
# for each that exceeded a threshold.
#
# WARNING: This adds low-overhead instrumentation to these ZFS operations,
# including reads and writes from the file system cache. Such reads and writes
# can be very frequent (depending on the workload; eg, 1M/sec), at which
# point the overhead of this tool (even if it prints no "slower" events) can
# begin to become significant.
#
# This works by using kernel dynamic tracing of the ZPL interface, and will
# need updates to match any changes to this interface.
#
# By default, a minimum millisecond threshold of 10 is used.
#
# Copyright 2016 Netflix, Inc.
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 14-Feb-2016 Brendan Gregg Created this.
# 16-Oct-2016 Dina Goldshtein -p to filter by process ID.
from __future__ import print_function
from bcc import BPF
import argparse
from time import strftime
import ctypes as ct
# arguments
examples = """examples:
./zfsslower # trace operations slower than 10 ms (default)
./zfsslower 1 # trace operations slower than 1 ms
./zfsslower -j 1 # ... 1 ms, parsable output (csv)
./zfsslower 0 # trace all operations (warning: verbose)
./zfsslower -p 185 # trace PID 185 only
"""
parser = argparse.ArgumentParser(
description="Trace common ZFS file operations slower than a threshold",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog=examples)
parser.add_argument("-j", "--csv", action="store_true",
help="just print fields: comma-separated values")
parser.add_argument("-p", "--pid",
help="trace this PID only")
parser.add_argument("min_ms", nargs="?", default='10',
help="minimum I/O duration to trace, in ms (default 10)")
parser.add_argument("--ebpf", action="store_true",
help=argparse.SUPPRESS)
args = parser.parse_args()
min_ms = int(args.min_ms)
pid = args.pid
csv = args.csv
debug = 0
# define BPF program
bpf_text = """
#include <uapi/linux/ptrace.h>
#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/dcache.h>
// XXX: switch these to char's when supported
#define TRACE_READ 0
#define TRACE_WRITE 1
#define TRACE_OPEN 2
#define TRACE_FSYNC 3
struct val_t {
u64 ts;
u64 offset;
struct file *fp;
};
struct data_t {
// XXX: switch some to u32's when supported
u64 ts_us;
u64 type;
u64 size;
u64 offset;
u64 delta_us;
u64 pid;
char task[TASK_COMM_LEN];
char file[DNAME_INLINE_LEN];
};
BPF_HASH(entryinfo, u64, struct val_t);
BPF_PERF_OUTPUT(events);
//
// Store timestamp and size on entry
//
// zpl_read(), zpl_write():
int trace_rw_entry(struct pt_regs *ctx, struct file *filp, char __user *buf,
size_t len, loff_t *ppos)
{
u64 id = bpf_get_current_pid_tgid();
u32 pid = id >> 32; // PID is higher part
if (FILTER_PID)
return 0;
// store filep and timestamp by id
struct val_t val = {};
val.ts = bpf_ktime_get_ns();
val.fp = filp;
val.offset = *ppos;
if (val.fp)
entryinfo.update(&id, &val);
return 0;
}
// zpl_open():
int trace_open_entry(struct pt_regs *ctx, struct inode *inode,
struct file *filp)
{
u64 id = bpf_get_current_pid_tgid();
u32 pid = id >> 32; // PID is higher part
if (FILTER_PID)
return 0;
// store filep and timestamp by id
struct val_t val = {};
val.ts = bpf_ktime_get_ns();
val.fp = filp;
val.offset = 0;
if (val.fp)
entryinfo.update(&id, &val);
return 0;
}
// zpl_fsync():
int trace_fsync_entry(struct pt_regs *ctx, struct file *filp)
{
u64 id = bpf_get_current_pid_tgid();
u32 pid = id >> 32; // PID is higher part
if (FILTER_PID)
return 0;
// store filp and timestamp by id
struct val_t val = {};
val.ts = bpf_ktime_get_ns();
val.fp = filp;
val.offset = 0;
if (val.fp)
entryinfo.update(&id, &val);
return 0;
}
//
// Output
//
static int trace_return(struct pt_regs *ctx, int type)
{
struct val_t *valp;
u64 id = bpf_get_current_pid_tgid();
u32 pid = id >> 32; // PID is higher part
valp = entryinfo.lookup(&id);
if (valp == 0) {
// missed tracing issue or filtered
return 0;
}
// calculate delta
u64 ts = bpf_ktime_get_ns();
u64 delta_us = (ts - valp->ts) / 1000;
entryinfo.delete(&id);
if (FILTER_US)
return 0;
// populate output struct
u32 size = PT_REGS_RC(ctx);
struct data_t data = {.type = type, .size = size, .delta_us = delta_us,
.pid = pid};
data.ts_us = ts / 1000;
data.offset = valp->offset;
bpf_get_current_comm(&data.task, sizeof(data.task));
struct qstr qs = valp->fp->f_path.dentry->d_name;
if (qs.len == 0)
return 0;
bpf_probe_read(&data.file, sizeof(data.file), (void *)qs.name);
// output
events.perf_submit(ctx, &data, sizeof(data));
return 0;
}
int trace_read_return(struct pt_regs *ctx)
{
return trace_return(ctx, TRACE_READ);
}
int trace_write_return(struct pt_regs *ctx)
{
return trace_return(ctx, TRACE_WRITE);
}
int trace_open_return(struct pt_regs *ctx)
{
return trace_return(ctx, TRACE_OPEN);
}
int trace_fsync_return(struct pt_regs *ctx)
{
return trace_return(ctx, TRACE_FSYNC);
}
"""
if min_ms == 0:
bpf_text = bpf_text.replace('FILTER_US', '0')
else:
bpf_text = bpf_text.replace('FILTER_US',
'delta_us <= %s' % str(min_ms * 1000))
if args.pid:
bpf_text = bpf_text.replace('FILTER_PID', 'pid != %s' % pid)
else:
bpf_text = bpf_text.replace('FILTER_PID', '0')
if debug or args.ebpf:
print(bpf_text)
if args.ebpf:
exit()
# kernel->user event data: struct data_t
DNAME_INLINE_LEN = 32 # linux/dcache.h
TASK_COMM_LEN = 16 # linux/sched.h
class Data(ct.Structure):
_fields_ = [
("ts_us", ct.c_ulonglong),
("type", ct.c_ulonglong),
("size", ct.c_ulonglong),
("offset", ct.c_ulonglong),
("delta_us", ct.c_ulonglong),
("pid", ct.c_ulonglong),
("task", ct.c_char * TASK_COMM_LEN),
("file", ct.c_char * DNAME_INLINE_LEN)
]
# process event
def print_event(cpu, data, size):
event = ct.cast(data, ct.POINTER(Data)).contents
type = 'R'
if event.type == 1:
type = 'W'
elif event.type == 2:
type = 'O'
elif event.type == 3:
type = 'S'
if (csv):
print("%d,%s,%d,%s,%d,%d,%d,%s" % (
event.ts_us, event.task.decode('utf-8', 'replace'), event.pid,
type, event.size, event.offset, event.delta_us,
event.file.decode('utf-8', 'replace')))
return
print("%-8s %-14.14s %-6s %1s %-7s %-8d %7.2f %s" % (strftime("%H:%M:%S"),
event.task.decode('utf-8', 'replace'), event.pid, type, event.size,
event.offset / 1024, float(event.delta_us) / 1000,
event.file.decode('utf-8', 'replace')))
# initialize BPF
b = BPF(text=bpf_text)
# common file functions
if BPF.get_kprobe_functions(b'zpl_iter'):
b.attach_kprobe(event="zpl_iter_read", fn_name="trace_rw_entry")
b.attach_kprobe(event="zpl_iter_write", fn_name="trace_rw_entry")
elif BPF.get_kprobe_functions(b'zpl_aio'):
b.attach_kprobe(event="zpl_aio_read", fn_name="trace_rw_entry")
b.attach_kprobe(event="zpl_aio_write", fn_name="trace_rw_entry")
else:
b.attach_kprobe(event="zpl_read", fn_name="trace_rw_entry")
b.attach_kprobe(event="zpl_write", fn_name="trace_rw_entry")
b.attach_kprobe(event="zpl_open", fn_name="trace_open_entry")
b.attach_kprobe(event="zpl_fsync", fn_name="trace_fsync_entry")
if BPF.get_kprobe_functions(b'zpl_iter'):
b.attach_kretprobe(event="zpl_iter_read", fn_name="trace_read_return")
b.attach_kretprobe(event="zpl_iter_write", fn_name="trace_write_return")
elif BPF.get_kprobe_functions(b'zpl_aio'):
b.attach_kretprobe(event="zpl_aio_read", fn_name="trace_read_return")
b.attach_kretprobe(event="zpl_aio_write", fn_name="trace_write_return")
else:
b.attach_kretprobe(event="zpl_read", fn_name="trace_read_return")
b.attach_kretprobe(event="zpl_write", fn_name="trace_write_return")
b.attach_kretprobe(event="zpl_open", fn_name="trace_open_return")
b.attach_kretprobe(event="zpl_fsync", fn_name="trace_fsync_return")
# header
if (csv):
print("ENDTIME_us,TASK,PID,TYPE,BYTES,OFFSET_b,LATENCY_us,FILE")
else:
if min_ms == 0:
print("Tracing ZFS operations")
else:
print("Tracing ZFS operations slower than %d ms" % min_ms)
print("%-8s %-14s %-6s %1s %-7s %-8s %7s %s" % ("TIME", "COMM", "PID", "T",
"BYTES", "OFF_KB", "LAT(ms)", "FILENAME"))
# read events
b["events"].open_perf_buffer(print_event, page_cnt=64)
while 1:
b.perf_buffer_poll()