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
|
# NVMe-oF Target Tracepoints {#nvmf_tgt_tracepoints}
# Introduction {#tracepoints_intro}
SPDK has a tracing framework for capturing low-level event information at runtime.
Tracepoints provide a high-performance tracing mechanism that is accessible at runtime.
They are implemented as a circular buffer in shared memory that is accessible from other
processes. The NVMe-oF target is instrumented with tracepoints to enable analysis of
both performance and application crashes. (Note: the SPDK tracing framework should still
be considered experimental. Work to formalize and document the framework is in progress.)
# Enabling Tracepoints {#enable_tracepoints}
Tracepoints are placed in groups. They are enabled and disabled as a group. To enable
the instrumentation of all the tracepoints group in an SPDK target application, start the
target with -e parameter set to 0xFFFF:
~~~
build/bin/nvmf_tgt -e 0xFFFF
~~~
To enable the instrumentation of just the NVMe-oF RDMA tracepoints in an SPDK target
application, start the target with the -e parameter set to 0x10:
~~~
build/bin/nvmf_tgt -e 0x10
~~~
When the target starts, a message is logged with the information you need to view
the tracepoints in a human-readable format using the spdk_trace application. The target
will also log information about the shared memory file.
~~~{.sh}
app.c: 527:spdk_app_setup_trace: *NOTICE*: Tracepoint Group Mask 0xFFFF specified.
app.c: 531:spdk_app_setup_trace: *NOTICE*: Use 'spdk_trace -s nvmf -p 24147' to capture a snapshot of events at runtime.
app.c: 533:spdk_app_setup_trace: *NOTICE*: Or copy /dev/shm/nvmf_trace.pid24147 for offline analysis/debug.
~~~
Note that when tracepoints are enabled, the shared memory files are not deleted when the application
exits. This ensures the file can be used for analysis after the application exits. On Linux, the
shared memory files are in /dev/shm, and can be deleted manually to free shm space if needed. A system
reboot will also free all of the /dev/shm files.
# Capturing a snapshot of events {#capture_tracepoints}
Send I/Os to the SPDK target application to generate events. The following is
an example usage of perf to send I/Os to the NVMe-oF target over an RDMA network
interface for 10 minutes.
~~~
./perf -q 128 -s 4096 -w randread -t 600 -r 'trtype:RDMA adrfam:IPv4 traddr:192.168.100.2 trsvcid:4420'
~~~
The spdk_trace program can be found in the app/trace directory. To analyze the tracepoints on the same
system running the NVMe-oF target, simply execute the command line shown in the log:
~~~{.sh}
build/bin/spdk_trace -s nvmf -p 24147
~~~
To analyze the tracepoints on a different system, first prepare the tracepoint file for transfer. The
tracepoint file can be large, but usually compresses very well. This step can also be used to prepare
a tracepoint file to attach to a GitHub issue for debugging NVMe-oF application crashes.
~~~{.sh}
bzip2 -c /dev/shm/nvmf_trace.pid24147 > /tmp/trace.bz2
~~~
After transferring the /tmp/trace.bz2 tracepoint file to a different system:
~~~{.sh}
bunzip2 /tmp/trace.bz2
build/bin/spdk_trace -f /tmp/trace
~~~
The following is sample trace capture showing the cumulative time that each
I/O spends at each RDMA state. All the trace captures with the same id are for
the same I/O.
~~~
28: 6026.658 ( 12656064) RDMA_REQ_NEED_BUFFER id: r3622 time: 0.019
28: 6026.694 ( 12656140) RDMA_REQ_RDY_TO_EXECUTE id: r3622 time: 0.055
28: 6026.820 ( 12656406) RDMA_REQ_EXECUTING id: r3622 time: 0.182
28: 6026.992 ( 12656766) RDMA_REQ_EXECUTED id: r3477 time: 228.510
28: 6027.010 ( 12656804) RDMA_REQ_TX_PENDING_C_TO_H id: r3477 time: 228.528
28: 6027.022 ( 12656828) RDMA_REQ_RDY_TO_COMPLETE id: r3477 time: 228.539
28: 6027.115 ( 12657024) RDMA_REQ_COMPLETING id: r3477 time: 228.633
28: 6027.471 ( 12657770) RDMA_REQ_COMPLETED id: r3518 time: 171.577
28: 6028.027 ( 12658940) RDMA_REQ_NEW id: r3623
28: 6028.057 ( 12659002) RDMA_REQ_NEED_BUFFER id: r3623 time: 0.030
28: 6028.095 ( 12659082) RDMA_REQ_RDY_TO_EXECUTE id: r3623 time: 0.068
28: 6028.216 ( 12659336) RDMA_REQ_EXECUTING id: r3623 time: 0.189
28: 6028.408 ( 12659740) RDMA_REQ_EXECUTED id: r3505 time: 190.509
28: 6028.441 ( 12659808) RDMA_REQ_TX_PENDING_C_TO_H id: r3505 time: 190.542
28: 6028.452 ( 12659832) RDMA_REQ_RDY_TO_COMPLETE id: r3505 time: 190.553
28: 6028.536 ( 12660008) RDMA_REQ_COMPLETING id: r3505 time: 190.637
28: 6028.854 ( 12660676) RDMA_REQ_COMPLETED id: r3465 time: 247.000
28: 6029.433 ( 12661892) RDMA_REQ_NEW id: r3624
28: 6029.452 ( 12661932) RDMA_REQ_NEED_BUFFER id: r3624 time: 0.019
28: 6029.482 ( 12661996) RDMA_REQ_RDY_TO_EXECUTE id: r3624 time: 0.050
28: 6029.591 ( 12662224) RDMA_REQ_EXECUTING id: r3624 time: 0.158
28: 6029.782 ( 12662624) RDMA_REQ_EXECUTED id: r3564 time: 96.937
28: 6029.798 ( 12662658) RDMA_REQ_TX_PENDING_C_TO_H id: r3564 time: 96.953
28: 6029.812 ( 12662688) RDMA_REQ_RDY_TO_COMPLETE id: r3564 time: 96.967
28: 6029.899 ( 12662870) RDMA_REQ_COMPLETING id: r3564 time: 97.054
28: 6030.262 ( 12663634) RDMA_REQ_COMPLETED id: r3477 time: 231.780
28: 6030.786 ( 12664734) RDMA_REQ_NEW id: r3625
28: 6030.804 ( 12664772) RDMA_REQ_NEED_BUFFER id: r3625 time: 0.018
28: 6030.841 ( 12664848) RDMA_REQ_RDY_TO_EXECUTE id: r3625 time: 0.054
28: 6030.963 ( 12665104) RDMA_REQ_EXECUTING id: r3625 time: 0.176
28: 6031.139 ( 12665474) RDMA_REQ_EXECUTED id: r3552 time: 114.906
28: 6031.196 ( 12665594) RDMA_REQ_TX_PENDING_C_TO_H id: r3552 time: 114.963
28: 6031.210 ( 12665624) RDMA_REQ_RDY_TO_COMPLETE id: r3552 time: 114.977
28: 6031.293 ( 12665798) RDMA_REQ_COMPLETING id: r3552 time: 115.060
28: 6031.633 ( 12666512) RDMA_REQ_COMPLETED id: r3505 time: 193.734
28: 6032.230 ( 12667766) RDMA_REQ_NEW id: r3626
28: 6032.248 ( 12667804) RDMA_REQ_NEED_BUFFER id: r3626 time: 0.018
28: 6032.288 ( 12667888) RDMA_REQ_RDY_TO_EXECUTE id: r3626 time: 0.058
28: 6032.396 ( 12668114) RDMA_REQ_EXECUTING id: r3626 time: 0.166
28: 6032.593 ( 12668528) RDMA_REQ_EXECUTED id: r3570 time: 90.443
28: 6032.611 ( 12668564) RDMA_REQ_TX_PENDING_C_TO_H id: r3570 time: 90.460
28: 6032.623 ( 12668590) RDMA_REQ_RDY_TO_COMPLETE id: r3570 time: 90.473
28: 6032.707 ( 12668766) RDMA_REQ_COMPLETING id: r3570 time: 90.557
28: 6033.056 ( 12669500) RDMA_REQ_COMPLETED id: r3564 time: 100.211
~~~
# Capturing sufficient trace events {#capture_trace_events}
Since the tracepoint file generated directly by SPDK application is a circular buffer in shared memory,
the trace events captured by it may be insufficient for further analysis.
The spdk_trace_record program can be found in the app/trace_record directory.
spdk_trace_record is used to poll the spdk tracepoint shared memory, record new entries from it,
and store all entries into specified output file at its shutdown on SIGINT or SIGTERM.
After SPDK nvmf target is launched, simply execute the command line shown in the log:
~~~{.sh}
build/bin/spdk_trace_record -q -s nvmf -p 24147 -f /tmp/spdk_nvmf_record.trace
~~~
Also send I/Os to the SPDK target application to generate events by previous perf example for 10 minutes.
~~~{.sh}
./perf -q 128 -s 4096 -w randread -t 600 -r 'trtype:RDMA adrfam:IPv4 traddr:192.168.100.2 trsvcid:4420'
~~~
After the completion of perf example, shut down spdk_trace_record by signal SIGINT (Ctrl + C).
To analyze the tracepoints output file from spdk_trace_record, simply run spdk_trace program by:
~~~{.sh}
build/bin/spdk_trace -f /tmp/spdk_nvmf_record.trace
~~~
# Adding New Tracepoints {#add_tracepoints}
SPDK applications and libraries provide several trace points. You can add new
tracepoints to the existing trace groups. For example, to add a new tracepoints
to the SPDK RDMA library (lib/nvmf/rdma.c) trace group TRACE_GROUP_NVMF_RDMA,
define the tracepoints and assigning them a unique ID using the SPDK_TPOINT_ID macro:
~~~
#define TRACE_GROUP_NVMF_RDMA 0x4
#define TRACE_RDMA_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_RDMA, 0x0)
...
#define NEW_TRACE_POINT_NAME SPDK_TPOINT_ID(TRACE_GROUP_NVMF_RDMA, UNIQUE_ID)
~~~
You also need to register the new trace points in the SPDK_TRACE_REGISTER_FN macro call
within the application/library using the spdk_trace_register_description function
as shown below:
~~~
SPDK_TRACE_REGISTER_FN(nvmf_trace)
{
spdk_trace_register_object(OBJECT_NVMF_RDMA_IO, 'r');
spdk_trace_register_description("RDMA_REQ_NEW", "",
TRACE_RDMA_REQUEST_STATE_NEW,
OWNER_NONE, OBJECT_NVMF_RDMA_IO, 1, 1, "cmid: ");
...
spdk_trace_register_description("NEW_RDMA_REQ_NAME", "",
NEW_TRACE_POINT_NAME,
OWNER_NONE, OBJECT_NVMF_RDMA_IO, 0, 1, "cmid: ");
}
~~~
Finally, use the spdk_trace_record function at the appropriate point in the
application/library to record the current trace state for the new trace points.
The following example shows the usage of the spdk_trace_record function to
record the current trace state of several tracepoints.
~~~
case RDMA_REQUEST_STATE_NEW:
spdk_trace_record(TRACE_RDMA_REQUEST_STATE_NEW, 0, 0, (uintptr_t)rdma_req, (uintptr_t)rqpair->cm_id);
...
break;
case RDMA_REQUEST_STATE_NEED_BUFFER:
spdk_trace_record(TRACE_RDMA_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)rdma_req, (uintptr_t)rqpair->cm_id);
...
break;
case RDMA_REQUEST_STATE_TRANSFER_PENDING_HOST_TO_CONTROLLER:
spdk_trace_record(TRACE_RDMA_REQUEST_STATE_TRANSFER_PENDING_HOST_TO_CONTROLLER, 0, 0,
(uintptr_t)rdma_req, (uintptr_t)rqpair->cm_id);
...
~~~
All the tracing functions are documented in the [Tracepoint library documentation](https://www.spdk.io/doc/trace_8h.html)
|
