1compactsnoop(8)             System Manager's Manual            compactsnoop(8)
2
3
4

NAME

6       compactstall - Trace compact zone events. Uses Linux eBPF/bcc.
7

SYNOPSIS

9       compactsnoop.py [-h] [-T] [-p PID] [-d DURATION] [-K] [-e]
10

DESCRIPTION

12       compactsnoop  traces  the  compact zone events, showing which processes
13       are allocing pages with memory compaction. This can be useful for  dis‐
14       covering  when  compact_stall  (/proc/vmstat)  continues  to  increase,
15       whether it is caused by some critical processes or not.
16
17       This works by tracing the compact zone events using raw_tracepoints and
18       one kretprobe.
19
20       For  the  Centos  7.6 (3.10.x kernel), see the version under tools/old,
21       which uses an older memory compaction mechanism.
22
23       Since this uses BPF, only the root user can use this tool.
24

REQUIREMENTS

26       CONFIG_BPF and bcc.
27

OPTIONS

29       -h     Print usage message.
30
31       -T     Include a timestamp column.
32
33       -p PID Trace this process ID only (filtered in-kernel).
34
35       -d DURATION
36              Total duration of trace in seconds.
37
38       -K     Output kernel stack trace
39
40       -e     Show extended fields.
41

EXAMPLES

43       Trace all compact zone events:
44              # compactsnoop
45
46       Trace all compact zone events, for 10 seconds only:
47              # compactsnoop -d 10
48

FIELDS

50       TIME(s)
51              Time of the call, in seconds.
52
53       COMM   Process name
54
55       PID    Process ID
56
57       NODE   Memory node
58
59       ZONE   Zone of the node (such as DMA, DMA32, NORMAL eg)
60
61       ORDER  Shows which order alloc cause memory compaction,  -1  means  all
62              orders (eg: write to /proc/sys/vm/compact_memory)
63
64       MODE   SYNC OR ASYNC
65
66       FRAGIDX (extra column)
67              The  FRAGIDX  is short for fragmentation index, which only makes
68              sense if an allocation of a requested size would fail.  If  that
69              is  true,  the  fragmentation  index  indicates whether external
70              fragmentation or a lack of memory was the problem. The value can
71              be  used  to  determine  if page reclaim or compaction should be
72              used.
73
74               Index is between 0 and 1 so return within 3 decimal places
75
76               0 => allocation would fail due to lack of memory
77
78               1 => allocation would fail due to fragmentation
79
80       MIN (extra column)
81              The min watermark of the zone
82
83       LOW (extra column)
84              The low watermark of the zone
85
86       HIGH (extra column)
87              The high watermark of the zone
88
89       FREE (extra column)
90              The nr_free_pages of the zone
91
92       LAT(ms)
93              compact zone's latency
94
95       STATUS The compaction's result.
96
97               For (CentOS 7.6's kernel), the status include:
98
99               "skipped" (COMPACT_SKIPPED): compaction didn't start as it  was
100               not possible or direct reclaim was more suitable
101
102               "continue"  (COMPACT_CONTINUE):  compaction  should continue to
103               another pageblock
104
105               "partial" (COMPACT_PARTIAL): direct compaction  partially  com‐
106               pacted a zone and there are suitable pages
107
108               "complete" (COMPACT_COMPLETE): The full zone was compacted
109
110               For (kernel 4.7 and above):
111
112               "not_suitable_zone"   (COMPACT_NOT_SUITABLE_ZONE):   For   more
113               detailed tracepoint output - internal to compaction
114
115               "skipped" (COMPACT_SKIPPED): compaction didn't start as it  was
116               not possible or direct reclaim was more suitable
117
118               "deferred"  (COMPACT_DEFERRED):  compaction  didn't start as it
119               was deferred due to past failures
120
121               "no_suitable_page"   (COMPACT_NOT_SUITABLE_PAGE):   For    more
122               detailed tracepoint output - internal to compaction
123
124               "continue"  (COMPACT_CONTINUE):  compaction  should continue to
125               another pageblock
126
127               "complete" (COMPACT_COMPLETE):  The  full  zone  was  compacted
128               scanned but wasn't successful to compact suitable pages.
129
130               "partial_skipped"  (COMPACT_PARTIAL_SKIPPED): direct compaction
131               has scanned part of the zone but wasn't successful  to  compact
132               suitable pages.
133
134               "contended"  (COMPACT_CONTENDED):  compaction terminated prema‐
135               turely due to lock contentions
136
137               "success" (COMPACT_SUCCESS): direct compaction terminated after
138               concluding that the allocation should now succeed
139

OVERHEAD

141       This traces the kernel compact zone kprobe/kretprobe or raw_tracepoints
142       and prints output for each event. As the  rate  of  this  is  generally
143       expected to be low (< 1000/s), the overhead is also expected to be neg‐
144       ligible.
145

SOURCE

147       This is from bcc.
148
149              https://github.com/iovisor/bcc
150
151       Also look in the bcc distribution for a  companion  _examples.txt  file
152       containing example usage, output, and commentary for this tool.
153

OS

155       Linux
156

STABILITY

158       Unstable - in development.
159

AUTHOR

161       Wenbo Zhang
162
163
164
165USER COMMANDS                      2019-11-1                   compactsnoop(8)