How to Debug Anything - DevOpsDay PGH

Transcript

1. How to Debug Anything @samkottler 1
2. About me • Engineering manager @ DigitalOcean • Formerly of Red Hat’s virtualization team • Ruby core security • CentOS/Icinga/Bundler/Rubygems/Ansible 2
3. tl;dr I professionally debug software and systems. 3
4. Most software is pretty bad. 4
5. Operating systems are no exception. 5
6. Verifying correctness on changing systems is impossible. 6
7. Turtles all the way down: • Unreliable hardware • Unreliable devices • Unreliable kernels • Unreliable drivers • Unreliable libraries • Unreliable runtimes 7
9. Bugs happen. Tools help quickly find and fix them. 9
10. But first, let's talk process. 10
11. 1. Suspend your disbelief "This isn't even possible!" 11
12. 2. Reproduce the issue Without being able to create the bad condition, it's impossible to fix it. 12
13. 3. Compare with healthy systems Sometimes simple or seemingly insignificant changes can lead to major issues. 13
14. 4. Figure out what's wrong. This is probably gonna take a while. 14
15. 5. Fix the problem Almost always easier than figuring out the problem because of the understanding built while debugging. 15
16. Let's jump back to step #4 16
17. Bugs happen. Tools help quickly find and fix them. 17
18. require "otherthing" module Derp class Yerp def self.string "derping in the USA" end end end puts Derp::Yerp.string 18
19. `require': cannot load such file -- otherthing (LoadError) from /usr/lib/ruby/1.9.1/rubygems/custom_require.rb:36:in `require' from test.rb:1:in `<main>' 19
20. open("/usr/lib/ruby/1.9.1/rubygems/path_support.rb", O_RDONLY) = 5 fstat(5, {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 fstat(5, {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 close(5) = 0 getuid() = 10100 geteuid() = 10100 getgid() = 10000 getegid() = 10000 open("/usr/lib/ruby/1.9.1/rubygems/path_support.rb", O_RDONLY) = 5 fstat(5, {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 ioctl(5, SNDCTL_TMR_TIMEBASE or SNDRV_TIMER_IOCTL_NEXT_DEVICE or TCGETS, 0x7ffc21fe8df0) = -1 ENOTTY (Inappropriate ioctl for device) fstat(5, {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 read(5, "##\n# Gem::PathSupport facilitate"..., 8192) = 1538 read(5, "", 8192) = 0 close(5) = 0 lstat("/usr", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat("/usr/lib", {st_mode=S_IFDIR|0755, st_size=12288, ...}) = 0 lstat("/usr/lib/ruby", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat("/usr/lib/ruby/1.9.1", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat("/usr/lib/ruby/1.9.1/rubygems", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 lstat("/usr/lib/ruby/1.9.1/rubygems/path_support.rb", {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 brk(0x1719000) = 0x1719000 stat("/home/skottler", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 openat(AT_FDCWD, "/var/lib/gems/1.9.1/specifications", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "/home/skottler/.gem/ruby/1.9.1/specifications", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = -1 ENOENT (No such file or directory) rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 open("/proc/self/maps", O_RDONLY|O_CLOEXEC) = 5 20
21. Reading strace output is easy! fstat(5, {st_mode=S_IFREG|0644, st_size=1538, ...}) = 0 21
22. Let's break it down: function(file descriptor, {....args.....}) = return value 22
23. System calls are a unified interface to debug userspace 23
24. Debugging a crash 24
25. #include <stdio.h> int main () { int foo[5], n; memset((char *)0x0, 1, 100); return 0; } 25
26. gcc -g -o foo foo.c 26
27. $ gdb foo 27
28. (gdb) list 1 #include <stdio.h> 2 3 int main () { 4 int foo[5], n; 5 6 memset((char *)0x0, 1, 100); 7 8 return 0; 9 } 28
29. (gdb) run Starting program: /home/skottler/foo Program received signal SIGSEGV, Segmentation fault. 0x00007ffff7aa1614 in memset () from /lib/x86_64-linux-gnu/libc.so.6 29
30. (gdb) where #0 0x00007ffff7aa1614 in memset () from /lib/x86_64-linux-gnu/libc.so.6 #1 0x0000000000400599 in main () at foo.c:6 30
31. (gdb) list 1 #include <stdio.h> 2 3 int main () { 4 int foo[5], n; 5 6 memset(&n, 0, sizeof(foo)); 7 8 return 0; 9 } 31
32. (gdb) run Starting program: /home/skottler/foo [Inferior 1 (process 44204) exited normally] 32
34. Profiling 34
35. The anatomy of a memory leak 35
36. A brief detour to redis-server 36
37. #include <stdlib.h> void f(void) { int* x = malloc(10 * sizeof(int)); x[10] = 0; } int main(void) { f(); return 0; } 37
38. gcc -g -o heap heap.c 38
39. valgrind --leak-check=yes heap 39
40. ==18809== Invalid write of size 4 ==18809== at 0x40054B: f (heap.c:5) ==18809== by 0x40055B: main (heap.c:9) ==18809== Address 0x51fc068 is 0 bytes after a block of size 40 alloc'd ==18809== at 0x4C2AB80: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==18809== by 0x40053E: f (heap.c:4) ==18809== by 0x40055B: main (heap.c:9) ==18809== ==18809== ==18809== HEAP SUMMARY: ==18809== in use at exit: 40 bytes in 1 blocks ==18809== total heap usage: 1 allocs, 0 frees, 40 bytes allocated ==18809== ==18809== 40 bytes in 1 blocks are definitely lost in loss record 1 of 1 ==18809== at 0x4C2AB80: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==18809== by 0x40053E: f (heap.c:4) ==18809== by 0x40055B: main (heap.c:9) 40
41. void f(void) { int* x = malloc(10 * sizeof(int)); x[10] = 0; } ==18809== Invalid write of size 4 ==18809== at 0x40054B: f (heap.c:5) ==18809== by 0x40055B: main (heap.c:9) 41
42. void f(void) { int* x = malloc(10 * sizeof(int)); x[9] = 0; } 42
43. void f(void) { int* x = malloc(10 * sizeof(int)); x[10] = 0; } ==18809== Address 0x51fc068 is 0 bytes after a block of size 40 alloc'd ==18809== at 0x4C2AB80: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==18809== by 0x40053E: f (heap.c:4) ==18809== by 0x40055B: main (heap.c:9) 43
44. void f(void) { int* x = malloc(10 * sizeof(int)); x[9] = 0; free(x); } 44
45. ==18855== HEAP SUMMARY: ==18855== in use at exit: 0 bytes in 0 blocks ==18855== total heap usage: 1 allocs, 1 frees, 40 bytes allocated ==18855== ==18855== All heap blocks were freed -- no leaks are possible ==18855== ==18855== For counts of detected and suppressed errors, rerun with: -v ==18855== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0) 45
46. 1. Suspend your disbelief 2. Reproduce the issue 3. Compare with healthy systems 4. Figure out what's wrong. 5. Fix the problem 46
47. Thanks! @samkottler sam@do.co github.com/skottler 47