This tutorial covers how to use bcc tools to quickly solve performance, troubleshooting, and networking issues. If you want to develop new bcc tools, see tutorial_bcc_python_developer.md for that tutorial.
It is assumed for this tutorial that bcc is already installed, and you can run tools like execsnoop successfully. See INSTALL.md. This uses enhancements added to the Linux 4.x series.
Some quick wins.
Before using bcc, you should start with the Linux basics. One reference is the Linux Performance Analysis in 60,000 Milliseconds post, which covers these commands:
- uptime
- dmesg | tail
- vmstat 1
- mpstat -P ALL 1
- pidstat 1
- iostat -xz 1
- free -m
- sar -n DEV 1
- sar -n TCP,ETCP 1
- top
Here is a generic checklist for performance investigations with bcc, first as a list, then in detail:
- execsnoop
- opensnoop
- ext4slower (or btrfs*, xfs*, zfs*)
- biolatency
- biosnoop
- cachestat
- tcpconnect
- tcpaccept
- tcpretrans
- runqlat
- profile
These tools may be installed on your system under /usr/share/bcc/tools, or you can run them from the bcc github repo under /tools where they have a .py extension. Browse the 50+ tools available for more analysis options.
# ./execsnoop
PCOMM PID RET ARGS
supervise 9660 0 ./run
supervise 9661 0 ./run
mkdir 9662 0 /bin/mkdir -p ./main
run 9663 0 ./run
[...]
execsnoop prints one line of output for each new process. Check for short-lived processes. These can consume CPU resources, but not show up in most monitoring tools that periodically take snapshots of which processes are running.
It works by tracing exec(), not the fork(), so it will catch many types of new processes but not all (eg, it won't see an application launching working processes, that doesn't exec() anything else).
More examples.
# ./opensnoop
PID COMM FD ERR PATH
1565 redis-server 5 0 /proc/1565/stat
1565 redis-server 5 0 /proc/1565/stat
1565 redis-server 5 0 /proc/1565/stat
1603 snmpd 9 0 /proc/net/dev
1603 snmpd 11 0 /proc/net/if_inet6
1603 snmpd -1 2 /sys/class/net/eth0/device/vendor
1603 snmpd 11 0 /proc/sys/net/ipv4/neigh/eth0/retrans_time_ms
1603 snmpd 11 0 /proc/sys/net/ipv6/neigh/eth0/retrans_time_ms
1603 snmpd 11 0 /proc/sys/net/ipv6/conf/eth0/forwarding
[...]
opensnoop prints one line of output for each open() syscall, including details.
Files that are opened can tell you a lot about how applications work: identifying their data files, config files, and log files. Sometimes applications can misbehave, and perform poorly, when they are constantly attempting to read files that do not exist. opensnoop gives you a quick look.
More examples.
# ./ext4slower
Tracing ext4 operations slower than 10 ms
TIME COMM PID T BYTES OFF_KB LAT(ms) FILENAME
06:35:01 cron 16464 R 1249 0 16.05 common-auth
06:35:01 cron 16463 R 1249 0 16.04 common-auth
06:35:01 cron 16465 R 1249 0 16.03 common-auth
06:35:01 cron 16465 R 4096 0 10.62 login.defs
06:35:01 cron 16464 R 4096 0 10.61 login.defs
ext4slower traces the ext4 file system and times common operations, and then only prints those that exceed a threshold.
This is great for identifying or exonerating one type of performance issue: show individually slow disk i/O via the file system. Disks process I/O asynchronously, and it can be difficult to associate latency at that layer with the latency applications experience. Tracing higher up in the kernel stack, at the VFS -> file system interface, will more closely match what an application suffers. Use this tool to identify if file system latency exceeds a given threshold.
Similar tools exist in bcc for other file systems: btrfsslower, xfsslower, and zfsslower. There is also fileslower, which works at the VFS layer and traces everything (although at some higher overhead).
More examples.
# ./biolatency
Tracing block device I/O... Hit Ctrl-C to end.
^C
usecs : count distribution
0 -> 1 : 0 | |
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 0 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 1 | |
128 -> 255 : 12 |******** |
256 -> 511 : 15 |********** |
512 -> 1023 : 43 |******************************* |
1024 -> 2047 : 52 |**************************************|
2048 -> 4095 : 47 |********************************** |
4096 -> 8191 : 52 |**************************************|
8192 -> 16383 : 36 |************************** |
16384 -> 32767 : 15 |********** |
32768 -> 65535 : 2 |* |
65536 -> 131071 : 2 |* |
biolatency traces disk I/O latency (time from device issue to completion), and when the tool ends (Ctrl-C, or a given interval), it prints a histogram summary of the latency.
This is great for understanding disk I/O latency beyond the average times given by tools like iostat. I/O latency outliers will be visible at the end of the distribution, as well as multi-mode distributions.
More examples.
# ./biosnoop
TIME(s) COMM PID DISK T SECTOR BYTES LAT(ms)
0.000004001 supervise 1950 xvda1 W 13092560 4096 0.74
0.000178002 supervise 1950 xvda1 W 13092432 4096 0.61
0.001469001 supervise 1956 xvda1 W 13092440 4096 1.24
0.001588002 supervise 1956 xvda1 W 13115128 4096 1.09
1.022346001 supervise 1950 xvda1 W 13115272 4096 0.98
1.022568002 supervise 1950 xvda1 W 13188496 4096 0.93
[...]
biosnoop prints a line of output for each disk I/O, with details including latency (time from device issue to completion).
This allows you to examine disk I/O in more detail, and look for time-ordered patterns (eg, reads queueing behind writes). Note that the output will be verbose if your system performs disk I/O at a high rate.
More examples.
# ./cachestat
HITS MISSES DIRTIES READ_HIT% WRITE_HIT% BUFFERS_MB CACHED_MB
1074 44 13 94.9% 2.9% 1 223
2195 170 8 92.5% 6.8% 1 143
182 53 56 53.6% 1.3% 1 143
62480 40960 20480 40.6% 19.8% 1 223
7 2 5 22.2% 22.2% 1 223
348 0 0 100.0% 0.0% 1 223
[...]
cachestat prints a one line summary every second (or every custom interval) showing statistics from the file system cache.
Use this to identify a low cache hit ratio, and a high rate of misses: which gives one lead for performance tuning.
More examples.
# ./tcpconnect
PID COMM IP SADDR DADDR DPORT
1479 telnet 4 127.0.0.1 127.0.0.1 23
1469 curl 4 10.201.219.236 54.245.105.25 80
1469 curl 4 10.201.219.236 54.67.101.145 80
1991 telnet 6 ::1 ::1 23
2015 ssh 6 fe80::2000:bff:fe82:3ac fe80::2000:bff:fe82:3ac 22
[...]
tcpconnect prints one line of output for every active TCP connection (eg, via connect()), with details including source and destination addresses.
Look for unexpected connections that may point to inefficiencies in application configuration, or an intruder.
More examples.
# ./tcpaccept
PID COMM IP RADDR LADDR LPORT
907 sshd 4 192.168.56.1 192.168.56.102 22
907 sshd 4 127.0.0.1 127.0.0.1 22
5389 perl 6 1234:ab12:2040:5020:2299:0:5:0 1234:ab12:2040:5020:2299:0:5:0 7001
[...]
tcpaccept prints one line of output for every passive TCP connection (eg, via accept()), with details including source and destination addresses.
Look for unexpected connections that may point to inefficiencies in application configuration, or an intruder.
More examples.
# ./tcpretrans
TIME PID IP LADDR:LPORT T> RADDR:RPORT STATE
01:55:05 0 4 10.153.223.157:22 R> 69.53.245.40:34619 ESTABLISHED
01:55:05 0 4 10.153.223.157:22 R> 69.53.245.40:34619 ESTABLISHED
01:55:17 0 4 10.153.223.157:22 R> 69.53.245.40:22957 ESTABLISHED
[...]
tcprerans prints one line of output for every TCP retransmit packet, with details including source and destination addresses, and kernel state of the TCP connection.
TCP retransmissions cause latency and throughput issues. For ESTABLISHED retransmits, look for patterns with networks. For SYN_SENT, this may point to target kernel CPU saturation and kernel packet drops.
More examples.
# ./runqlat
Tracing run queue latency... Hit Ctrl-C to end.
^C
usecs : count distribution
0 -> 1 : 233 |*********** |
2 -> 3 : 742 |************************************ |
4 -> 7 : 203 |********** |
8 -> 15 : 173 |******** |
16 -> 31 : 24 |* |
32 -> 63 : 0 | |
64 -> 127 : 30 |* |
128 -> 255 : 6 | |
256 -> 511 : 3 | |
512 -> 1023 : 5 | |
1024 -> 2047 : 27 |* |
2048 -> 4095 : 30 |* |
4096 -> 8191 : 20 | |
8192 -> 16383 : 29 |* |
16384 -> 32767 : 809 |****************************************|
32768 -> 65535 : 64 |*** |
runqlat times how long threads were waiting on the CPU run queues, and prints this as a histogram.
This can help quantify time lost waiting for a turn on CPU, during periods of CPU saturation.
More examples.
# ./profile
Sampling at 49 Hertz of all threads by user + kernel stack... Hit Ctrl-C to end.
^C
00007f31d76c3251 [unknown]
47a2c1e752bf47f7 [unknown]
- sign-file (8877)
1
ffffffff813d0af8 __clear_user
ffffffff813d5277 iov_iter_zero
ffffffff814ec5f2 read_iter_zero
ffffffff8120be9d __vfs_read
ffffffff8120c385 vfs_read
ffffffff8120d786 sys_read
ffffffff817cc076 entry_SYSCALL_64_fastpath
00007fc5652ad9b0 read
- dd (25036)
4
0000000000400542 func_a
0000000000400598 main
00007f12a133e830 __libc_start_main
083e258d4c544155 [unknown]
- func_ab (13549)
5
[...]
ffffffff8105eb66 native_safe_halt
ffffffff8103659e default_idle
ffffffff81036d1f arch_cpu_idle
ffffffff810bba5a default_idle_call
ffffffff810bbd07 cpu_startup_entry
ffffffff8104df55 start_secondary
- swapper/1 (0)
75
profile is a CPU profiler, which takes samples of stack traces at timed intervals, and prints a summary of unique stack traces and a count of their occurrence.
Use this tool to understand the code paths that are consuming CPU resources.
More examples.
In addition to the above tools for performance tuning, below is a checklist for bcc generic tools, first as a list, and in detail:
- trace
- argdist
- funccount
These generic tools may be useful to provide visibility to solve your specific problems.
Suppose you want to track file ownership change. There are three syscalls, chown, fchown and lchown which users can use to change file ownership. The corresponding syscall entry is SyS_[f|l]chown. The following command can be used to print out syscall parameters and the calling process user id. You can use id command to find the uid of a particular user.
$ trace.py \
'p::SyS_chown "file = %s, to_uid = %d, to_gid = %d, from_uid = %d", arg1, arg2, arg3, $uid' \
'p::SyS_fchown "fd = %d, to_uid = %d, to_gid = %d, from_uid = %d", arg1, arg2, arg3, $uid' \
'p::SyS_lchown "file = %s, to_uid = %d, to_gid = %d, from_uid = %d", arg1, arg2, arg3, $uid'
PID TID COMM FUNC -
1269255 1269255 python3.6 SyS_lchown file = /tmp/dotsync-usisgezu/tmp, to_uid = 128203, to_gid = 100, from_uid = 128203
1269441 1269441 zstd SyS_chown file = /tmp/dotsync-vic7ygj0/dotsync-package.zst, to_uid = 128203, to_gid = 100, from_uid = 128203
1269255 1269255 python3.6 SyS_lchown file = /tmp/dotsync-a40zd7ev/tmp, to_uid = 128203, to_gid = 100, from_uid = 128203
1269442 1269442 zstd SyS_chown file = /tmp/dotsync-gzp413o_/dotsync-package.zst, to_uid = 128203, to_gid = 100, from_uid = 128203
1269255 1269255 python3.6 SyS_lchown file = /tmp/dotsync-whx4fivm/tmp/.bash_profile, to_uid = 128203, to_gid = 100, from_uid = 128203
Suppose you want to count nonvoluntary context switches (nvcsw) in your bpf based performance monitoring tools and you do not know what is the proper method. /proc/<pid>/status already tells you the number (nonvoluntary_ctxt_switches) for a pid and you can use trace.py to do a quick experiment to verify your method. With kernel source code, the nvcsw is counted at file linux/kernel/sched/core.c function __schedule and under condition
!(!preempt && prev->state) // i.e., preempt || !prev->state
The __schedule function is marked as notrace, and the best place to evaluate the above condition seems in sched/sched_switch tracepoint called inside function __schedule and defined in linux/include/trace/events/sched.h. trace.py already has args being the pointer to the tracepoint TP_STRUCT__entry. The above condition in function __schedule can be represented as
args->prev_state == TASK_STATE_MAX || args->prev_state == 0
The below command can be used to count the involuntary context switches (per process or per pid) and compare to /proc/<pid>/status or /proc/<pid>/task/<task_id>/status for correctness, as in typical cases, involuntary context switches are not very common.
$ trace.py -p 1134138 't:sched:sched_switch (args->prev_state == TASK_STATE_MAX || args->prev_state == 0)'
PID TID COMM FUNC
1134138 1134140 contention_test sched_switch
1134138 1134142 contention_test sched_switch
...
$ trace.py -L 1134140 't:sched:sched_switch (args->prev_state == TASK_STATE_MAX || args->prev_state == 0)'
PID TID COMM FUNC
1134138 1134140 contention_test sched_switch
1134138 1134140 contention_test sched_switch
...
This example is related to issue 1231 and 1516 where uprobe does not work at all in certain cases. First, you can do a strace as below
$ strace trace.py 'r:bash:readline "%s", retval'
...
perf_event_open(0x7ffd968212f0, -1, 0, -1, 0x8 /* PERF_FLAG_??? */) = -1 EIO (Input/output error)
...
The perf_event_open syscall returns -EIO. Digging into kernel uprobe related codes in /kernel/trace and /kernel/events directories to search EIO, the function uprobe_register is the most suspicious. Let us find whether this function is called or not and what is the return value if it is called. In one terminal using the following command to print out the return value of uprobe_register,
$ trace.py 'r::uprobe_register "ret = %d", retval'
In another terminal run the same bash uretprobe tracing example, and you should get
$ trace.py 'r::uprobe_register "ret = %d", retval'
PID TID COMM FUNC -
1041401 1041401 python2.7 uprobe_register ret = -5
The -5 error code is EIO. This confirms that the following code in function uprobe_register is the most suspicious culprit.
if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
return -EIO;
The shmem_mapping function is defined as
bool shmem_mapping(struct address_space *mapping)
{
return mapping->a_ops == &shmem_aops;
}
To confirm the theory, find what is inode->i_mapping->a_ops with the following command
$ trace.py -I 'linux/fs.h' 'p::uprobe_register(struct inode *inode) "a_ops = %llx", inode->i_mapping->a_ops'
PID TID COMM FUNC -
814288 814288 python2.7 uprobe_register a_ops = ffffffff81a2adc0
^C$ grep ffffffff81a2adc0 /proc/kallsyms
ffffffff81a2adc0 R empty_aops
The kernel symbol empty_aops does not have readpage defined and hence the above suspicious condition is true. Further examining the kernel source code shows that overlayfs does not provide its own a_ops while some other file systems (e.g., ext4) define their own a_ops (e.g., ext4_da_aops), and ext4_da_aops defines readpage. Hence, uprobe works fine on ext4 while not on overlayfs.
More examples.
More examples.
More examples.
To do.