prometheus-client-mmap 0.9.1 → 0.9.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/ext/fast_mmaped_file/#file_parsing.c# +216 -0
- data/ext/fast_mmaped_file/#value_access.c# +203 -0
- data/lib/prometheus/client.rb +5 -0
- data/lib/prometheus/client/#histogram.rb# +79 -0
- data/lib/prometheus/client/#mmaped_dict.rb# +108 -0
- data/lib/prometheus/client/helper/#entry_parser.rb# +132 -0
- data/lib/prometheus/client/version.rb +1 -1
- metadata +8 -7
- data/ext/fast_mmaped_file/hashmap.c +0 -692
- data/ext/fast_mmaped_file/jsmn.c +0 -314
- data/lib/fast_mmaped_file.bundle +0 -0
- data/vendor/c/jsmn/jsmn.o +0 -0
@@ -0,0 +1,108 @@
|
|
1
|
+
require 'prometheus/client/helper/mmaped_file'
|
2
|
+
require 'prometheus/client'
|
3
|
+
|
4
|
+
module Prometheus
|
5
|
+
module Client
|
6
|
+
class ParsingError < StandardError
|
7
|
+
end
|
8
|
+
|
9
|
+
# A dict of doubles, backed by an mmapped file.
|
10
|
+
#
|
11
|
+
# The file starts with a 4 byte int, indicating how much of it is used.
|
12
|
+
# Then 4 bytes of padding.
|
13
|
+
# There's then a number of entries, consisting of a 4 byte int which is the
|
14
|
+
# size of the next field, a utf-8 encoded string key, padding to an 8 byte
|
15
|
+
# alignment, and then a 8 byte float which is the value.
|
16
|
+
#
|
17
|
+
# For example, for :
|
18
|
+
#
|
19
|
+
# 0: | PID (4 bytes) | Used bytes (4 bytes) |
|
20
|
+
# 1: | Padding (4 bytes) | Size of field (4 bytes) |
|
21
|
+
# 2:
|
22
|
+
# 2: | Entry 1 key (N bytes) | Padding (4 bytes) |
|
23
|
+
# 3: | Entry 1 value (8 bytes) |
|
24
|
+
# 4: | Size of field 2 | Padding (4 bytes) |
|
25
|
+
# 5: | Entry 2 key (4 bytes) | Padding (4 byte) |
|
26
|
+
# 6: | Entry 2 value (8 bytes) |
|
27
|
+
class MmapedDict
|
28
|
+
MINIMUM_SIZE = 8
|
29
|
+
attr_reader :m, :used, :positions
|
30
|
+
|
31
|
+
def initialize(m)
|
32
|
+
@mutex = Mutex.new
|
33
|
+
|
34
|
+
@m = m
|
35
|
+
# @m.mlock # TODO: Ensure memory is locked to RAM
|
36
|
+
|
37
|
+
@used = @m.used
|
38
|
+
|
39
|
+
@positions = {}
|
40
|
+
read_all_positions.each do |key, pos|
|
41
|
+
@positions[key] = pos
|
42
|
+
end
|
43
|
+
rescue StandardError => e
|
44
|
+
raise ParsingError, "exception #{e} while processing metrics file #{path}"
|
45
|
+
end
|
46
|
+
|
47
|
+
# Yield (key, value). No locking is performed.
|
48
|
+
def self.read_all_values(f)
|
49
|
+
m = Helper::MmapedFile.open(f)
|
50
|
+
|
51
|
+
m.entries.map do |data, encoded_len, value_offset, _|
|
52
|
+
encoded, value = data.unpack(format('@4A%d@%dd', encoded_len, value_offset))
|
53
|
+
[encoded, value]
|
54
|
+
end
|
55
|
+
ensure
|
56
|
+
m.munmap
|
57
|
+
end
|
58
|
+
|
59
|
+
def read_value(key)
|
60
|
+
@mutex.synchronize do
|
61
|
+
init_value(key) unless @positions.key?(key)
|
62
|
+
end
|
63
|
+
pos = @positions[key]
|
64
|
+
# We assume that reading from an 8 byte aligned value is atomic.
|
65
|
+
@m[pos..pos + 7].unpack('d')[0]
|
66
|
+
end
|
67
|
+
|
68
|
+
def write_value(key, value)
|
69
|
+
@mutex.synchronize do
|
70
|
+
init_value(key) unless @positions.key?(key)
|
71
|
+
end
|
72
|
+
pos = @positions[key]
|
73
|
+
# We assume that writing to an 8 byte aligned value is atomic.
|
74
|
+
@m[pos..pos + 7] = [value].pack('d')
|
75
|
+
end
|
76
|
+
|
77
|
+
def path
|
78
|
+
@m.filepath unless @m.nil?
|
79
|
+
end
|
80
|
+
|
81
|
+
def close
|
82
|
+
@m.close
|
83
|
+
rescue TypeError => e
|
84
|
+
Prometheus::Client.logger.warn("munmap raised error #{e}")
|
85
|
+
end
|
86
|
+
|
87
|
+
private
|
88
|
+
|
89
|
+
# Initialize a value. Lock must be held by caller.
|
90
|
+
def init_value(key)
|
91
|
+
@m.add_entry(key, 0.0)
|
92
|
+
|
93
|
+
# Update how much space we've used.
|
94
|
+
@used = @m.used
|
95
|
+
|
96
|
+
@positions[key] = @used - 8
|
97
|
+
end
|
98
|
+
|
99
|
+
# Yield (key, pos). No locking is performed.
|
100
|
+
def read_all_positions
|
101
|
+
@m.entries.map do |data, encoded_len, _, absolute_pos|
|
102
|
+
encoded, = data.unpack(format('@4A%d', encoded_len))
|
103
|
+
[encoded, absolute_pos]
|
104
|
+
end
|
105
|
+
end
|
106
|
+
end
|
107
|
+
end
|
108
|
+
end
|
@@ -0,0 +1,132 @@
|
|
1
|
+
require 'prometheus/client/helper/json_parser'
|
2
|
+
|
3
|
+
module Prometheus
|
4
|
+
module Client
|
5
|
+
module Helper
|
6
|
+
module EntryParser
|
7
|
+
class ParsingError < RuntimeError;
|
8
|
+
end
|
9
|
+
|
10
|
+
MINIMUM_SIZE = 8
|
11
|
+
START_POSITION = 8
|
12
|
+
VALUE_BYTES = 8
|
13
|
+
ENCODED_LENGTH_BYTES = 4
|
14
|
+
|
15
|
+
def used
|
16
|
+
slice(0..3).unpack('l')[0]
|
17
|
+
end
|
18
|
+
|
19
|
+
def parts
|
20
|
+
@parts ||= File.basename(filepath, '.db')
|
21
|
+
.split('_')
|
22
|
+
.map { |e| e.gsub(/-\d+$/, '') } # remove trailing -number
|
23
|
+
end
|
24
|
+
|
25
|
+
def type
|
26
|
+
parts[0].to_sym
|
27
|
+
end
|
28
|
+
|
29
|
+
def pid
|
30
|
+
parts[2..-1].join('_')
|
31
|
+
end
|
32
|
+
|
33
|
+
def multiprocess_mode
|
34
|
+
parts[1]
|
35
|
+
end
|
36
|
+
|
37
|
+
def empty?
|
38
|
+
size < MINIMUM_SIZE || used.zero?
|
39
|
+
end
|
40
|
+
|
41
|
+
def entries(ignore_errors = false)
|
42
|
+
return Enumerator.new {} if empty?
|
43
|
+
|
44
|
+
Enumerator.new do |yielder|
|
45
|
+
used_ = used # cache used to avoid unnecessary unpack operations
|
46
|
+
|
47
|
+
pos = START_POSITION # used + padding offset
|
48
|
+
while pos < used_ && pos < size && pos > 0
|
49
|
+
data = slice(pos..-1)
|
50
|
+
unless data
|
51
|
+
raise ParsingError, "data slice is nil at pos #{pos}" unless ignore_errors
|
52
|
+
pos += 8c
|
53
|
+
next
|
54
|
+
end
|
55
|
+
|
56
|
+
encoded_len, first_encoded_bytes = data.unpack('LL')
|
57
|
+
if encoded_len.nil? || encoded_len.zero? || first_encoded_bytes.nil? || first_encoded_bytes.zero?
|
58
|
+
# do not parse empty data
|
59
|
+
pos += 8
|
60
|
+
next
|
61
|
+
end
|
62
|
+
|
63
|
+
entry_len = ENCODED_LENGTH_BYTES + encoded_len
|
64
|
+
padding_len = 8 - entry_len % 8
|
65
|
+
|
66
|
+
value_offset = entry_len + padding_len # align to 8 bytes
|
67
|
+
pos += value_offset
|
68
|
+
|
69
|
+
if value_offset > 0 && (pos + VALUE_BYTES) < size # if positions are safe
|
70
|
+
yielder.yield data, encoded_len, value_offset, pos
|
71
|
+
else
|
72
|
+
raise ParsingError, "data slice is nil at pos #{pos}" unless ignore_errors
|
73
|
+
end
|
74
|
+
pos += VALUE_BYTES
|
75
|
+
end
|
76
|
+
end
|
77
|
+
end
|
78
|
+
|
79
|
+
def parsed_entries(ignore_errors = false)
|
80
|
+
result = entries(ignore_errors).map do |data, encoded_len, value_offset, _|
|
81
|
+
begin
|
82
|
+
encoded, value = data.unpack(format('@4A%d@%dd', encoded_len, value_offset))
|
83
|
+
[encoded, value]
|
84
|
+
rescue ArgumentError => e
|
85
|
+
Prometheus::Client.logger.debug("Error processing data: #{bin_to_hex(data[0, 7])} len: #{encoded_len} value_offset: #{value_offset}")
|
86
|
+
raise ParsingError, e unless ignore_errors
|
87
|
+
end
|
88
|
+
end
|
89
|
+
|
90
|
+
metrics.reject!(&:nil?) if ignore_errors
|
91
|
+
result
|
92
|
+
end
|
93
|
+
|
94
|
+
def to_metrics(metrics = {}, ignore_errors = false)
|
95
|
+
parsed_entries(ignore_errors).each do |key, value|
|
96
|
+
begin
|
97
|
+
metric_name, name, labelnames, labelvalues = JsonParser.load(key)
|
98
|
+
labelnames ||= []
|
99
|
+
labelvalues ||= []
|
100
|
+
|
101
|
+
metric = metrics.fetch(metric_name,
|
102
|
+
metric_name: metric_name,
|
103
|
+
help: 'Multiprocess metric',
|
104
|
+
type: type,
|
105
|
+
samples: [])
|
106
|
+
if type == :gauge
|
107
|
+
metric[:multiprocess_mode] = multiprocess_mode
|
108
|
+
metric[:samples] += [[name, labelnames.zip(labelvalues) + [['pid', pid]], value]]
|
109
|
+
else
|
110
|
+
# The duplicates and labels are fixed in the next for.
|
111
|
+
metric[:samples] += [[name, labelnames.zip(labelvalues), value]]
|
112
|
+
end
|
113
|
+
metrics[metric_name] = metric
|
114
|
+
|
115
|
+
rescue JSON::ParserError => e
|
116
|
+
raise ParsingError(e) unless ignore_errors
|
117
|
+
end
|
118
|
+
end
|
119
|
+
|
120
|
+
metrics.reject!(&:nil?) if ignore_errors
|
121
|
+
metrics
|
122
|
+
end
|
123
|
+
|
124
|
+
private
|
125
|
+
|
126
|
+
def bin_to_hex(s)
|
127
|
+
s.each_byte.map { |b| b.to_s(16) }.join
|
128
|
+
end
|
129
|
+
end
|
130
|
+
end
|
131
|
+
end
|
132
|
+
end
|
metadata
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: prometheus-client-mmap
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.9.
|
4
|
+
version: 0.9.2
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Tobias Schmidt
|
@@ -9,7 +9,7 @@ authors:
|
|
9
9
|
autorequire:
|
10
10
|
bindir: bin
|
11
11
|
cert_chain: []
|
12
|
-
date: 2018-
|
12
|
+
date: 2018-05-09 00:00:00.000000000 Z
|
13
13
|
dependencies:
|
14
14
|
- !ruby/object:Gem::Dependency
|
15
15
|
name: fuzzbert
|
@@ -83,6 +83,8 @@ extensions:
|
|
83
83
|
extra_rdoc_files: []
|
84
84
|
files:
|
85
85
|
- README.md
|
86
|
+
- ext/fast_mmaped_file/#file_parsing.c#
|
87
|
+
- ext/fast_mmaped_file/#value_access.c#
|
86
88
|
- ext/fast_mmaped_file/extconf.rb
|
87
89
|
- ext/fast_mmaped_file/fast_mmaped_file.c
|
88
90
|
- ext/fast_mmaped_file/file_format.c
|
@@ -92,8 +94,6 @@ files:
|
|
92
94
|
- ext/fast_mmaped_file/file_reading.c
|
93
95
|
- ext/fast_mmaped_file/file_reading.h
|
94
96
|
- ext/fast_mmaped_file/globals.h
|
95
|
-
- ext/fast_mmaped_file/hashmap.c
|
96
|
-
- ext/fast_mmaped_file/jsmn.c
|
97
97
|
- ext/fast_mmaped_file/mmap.c
|
98
98
|
- ext/fast_mmaped_file/mmap.h
|
99
99
|
- ext/fast_mmaped_file/rendering.c
|
@@ -102,13 +102,15 @@ files:
|
|
102
102
|
- ext/fast_mmaped_file/utils.h
|
103
103
|
- ext/fast_mmaped_file/value_access.c
|
104
104
|
- ext/fast_mmaped_file/value_access.h
|
105
|
-
- lib/fast_mmaped_file.bundle
|
106
105
|
- lib/prometheus.rb
|
107
106
|
- lib/prometheus/client.rb
|
107
|
+
- lib/prometheus/client/#histogram.rb#
|
108
|
+
- lib/prometheus/client/#mmaped_dict.rb#
|
108
109
|
- lib/prometheus/client/configuration.rb
|
109
110
|
- lib/prometheus/client/counter.rb
|
110
111
|
- lib/prometheus/client/formats/text.rb
|
111
112
|
- lib/prometheus/client/gauge.rb
|
113
|
+
- lib/prometheus/client/helper/#entry_parser.rb#
|
112
114
|
- lib/prometheus/client/helper/entry_parser.rb
|
113
115
|
- lib/prometheus/client/helper/file_locker.rb
|
114
116
|
- lib/prometheus/client/helper/json_parser.rb
|
@@ -144,7 +146,6 @@ files:
|
|
144
146
|
- vendor/c/jsmn/example/simple.c
|
145
147
|
- vendor/c/jsmn/jsmn.c
|
146
148
|
- vendor/c/jsmn/jsmn.h
|
147
|
-
- vendor/c/jsmn/jsmn.o
|
148
149
|
- vendor/c/jsmn/library.json
|
149
150
|
- vendor/c/jsmn/test/test.h
|
150
151
|
- vendor/c/jsmn/test/tests.c
|
@@ -169,7 +170,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
|
|
169
170
|
version: '0'
|
170
171
|
requirements: []
|
171
172
|
rubyforge_project:
|
172
|
-
rubygems_version: 2.
|
173
|
+
rubygems_version: 2.5.2.2
|
173
174
|
signing_key:
|
174
175
|
specification_version: 4
|
175
176
|
summary: A suite of instrumentation metric primitivesthat can be exposed through a
|
@@ -1,692 +0,0 @@
|
|
1
|
-
/*
|
2
|
-
* Copyright (c) 2016-2017 David Leeds <davidesleeds@gmail.com>
|
3
|
-
*
|
4
|
-
* Hashmap is free software; you can redistribute it and/or modify
|
5
|
-
* it under the terms of the MIT license. See LICENSE for details.
|
6
|
-
*/
|
7
|
-
|
8
|
-
#include <stdlib.h>
|
9
|
-
#include <stdint.h>
|
10
|
-
#include <stdbool.h>
|
11
|
-
#include <string.h>
|
12
|
-
#include <errno.h>
|
13
|
-
|
14
|
-
#include "hashmap.h"
|
15
|
-
|
16
|
-
#ifndef HASHMAP_NOASSERT
|
17
|
-
#include <assert.h>
|
18
|
-
#define HASHMAP_ASSERT(expr) assert(expr)
|
19
|
-
#else
|
20
|
-
#define HASHMAP_ASSERT(expr)
|
21
|
-
#endif
|
22
|
-
|
23
|
-
/* Table sizes must be powers of 2 */
|
24
|
-
#define HASHMAP_SIZE_MIN (1 << 5) /* 32 */
|
25
|
-
#define HASHMAP_SIZE_DEFAULT (1 << 8) /* 256 */
|
26
|
-
#define HASHMAP_SIZE_MOD(map, val) ((val) & ((map)->table_size - 1))
|
27
|
-
|
28
|
-
/* Limit for probing is 1/2 of table_size */
|
29
|
-
#define HASHMAP_PROBE_LEN(map) ((map)->table_size >> 1)
|
30
|
-
/* Return the next linear probe index */
|
31
|
-
#define HASHMAP_PROBE_NEXT(map, index) HASHMAP_SIZE_MOD(map, (index) + 1)
|
32
|
-
|
33
|
-
/* Check if index b is less than or equal to index a */
|
34
|
-
#define HASHMAP_INDEX_LE(map, a, b) \
|
35
|
-
((a) == (b) || (((b) - (a)) & ((map)->table_size >> 1)) != 0)
|
36
|
-
|
37
|
-
|
38
|
-
struct hashmap_entry {
|
39
|
-
void *key;
|
40
|
-
void *data;
|
41
|
-
#ifdef HASHMAP_METRICS
|
42
|
-
size_t num_collisions;
|
43
|
-
#endif
|
44
|
-
};
|
45
|
-
|
46
|
-
|
47
|
-
/*
|
48
|
-
* Enforce a maximum 0.75 load factor.
|
49
|
-
*/
|
50
|
-
static inline size_t hashmap_table_min_size_calc(size_t num_entries)
|
51
|
-
{
|
52
|
-
return num_entries + (num_entries / 3);
|
53
|
-
}
|
54
|
-
|
55
|
-
/*
|
56
|
-
* Calculate the optimal table size, given the specified max number
|
57
|
-
* of elements.
|
58
|
-
*/
|
59
|
-
static size_t hashmap_table_size_calc(size_t num_entries)
|
60
|
-
{
|
61
|
-
size_t table_size;
|
62
|
-
size_t min_size;
|
63
|
-
|
64
|
-
table_size = hashmap_table_min_size_calc(num_entries);
|
65
|
-
|
66
|
-
/* Table size is always a power of 2 */
|
67
|
-
min_size = HASHMAP_SIZE_MIN;
|
68
|
-
while (min_size < table_size) {
|
69
|
-
min_size <<= 1;
|
70
|
-
}
|
71
|
-
return min_size;
|
72
|
-
}
|
73
|
-
|
74
|
-
/*
|
75
|
-
* Get a valid hash table index from a key.
|
76
|
-
*/
|
77
|
-
static inline size_t hashmap_calc_index(const struct hashmap *map,
|
78
|
-
const void *key)
|
79
|
-
{
|
80
|
-
return HASHMAP_SIZE_MOD(map, map->hash(key));
|
81
|
-
}
|
82
|
-
|
83
|
-
/*
|
84
|
-
* Return the next populated entry, starting with the specified one.
|
85
|
-
* Returns NULL if there are no more valid entries.
|
86
|
-
*/
|
87
|
-
static struct hashmap_entry *hashmap_entry_get_populated(
|
88
|
-
const struct hashmap *map, struct hashmap_entry *entry)
|
89
|
-
{
|
90
|
-
for (; entry < &map->table[map->table_size]; ++entry) {
|
91
|
-
if (entry->key) {
|
92
|
-
return entry;
|
93
|
-
}
|
94
|
-
}
|
95
|
-
return NULL;
|
96
|
-
}
|
97
|
-
|
98
|
-
/*
|
99
|
-
* Find the hashmap entry with the specified key, or an empty slot.
|
100
|
-
* Returns NULL if the entire table has been searched without finding a match.
|
101
|
-
*/
|
102
|
-
static struct hashmap_entry *hashmap_entry_find(const struct hashmap *map,
|
103
|
-
const void *key, bool find_empty)
|
104
|
-
{
|
105
|
-
size_t i;
|
106
|
-
size_t index;
|
107
|
-
size_t probe_len = HASHMAP_PROBE_LEN(map);
|
108
|
-
struct hashmap_entry *entry;
|
109
|
-
|
110
|
-
index = hashmap_calc_index(map, key);
|
111
|
-
|
112
|
-
/* Linear probing */
|
113
|
-
for (i = 0; i < probe_len; ++i) {
|
114
|
-
entry = &map->table[index];
|
115
|
-
if (!entry->key) {
|
116
|
-
if (find_empty) {
|
117
|
-
#ifdef HASHMAP_METRICS
|
118
|
-
entry->num_collisions = i;
|
119
|
-
#endif
|
120
|
-
return entry;
|
121
|
-
}
|
122
|
-
return NULL;
|
123
|
-
}
|
124
|
-
if (map->key_compare(key, entry->key) == 0) {
|
125
|
-
return entry;
|
126
|
-
}
|
127
|
-
index = HASHMAP_PROBE_NEXT(map, index);
|
128
|
-
}
|
129
|
-
return NULL;
|
130
|
-
}
|
131
|
-
|
132
|
-
/*
|
133
|
-
* Removes the specified entry and processes the proceeding entries to reduce
|
134
|
-
* the load factor and keep the chain continuous. This is a required
|
135
|
-
* step for hash maps using linear probing.
|
136
|
-
*/
|
137
|
-
static void hashmap_entry_remove(struct hashmap *map,
|
138
|
-
struct hashmap_entry *removed_entry)
|
139
|
-
{
|
140
|
-
size_t i;
|
141
|
-
#ifdef HASHMAP_METRICS
|
142
|
-
size_t removed_i = 0;
|
143
|
-
#endif
|
144
|
-
size_t index;
|
145
|
-
size_t entry_index;
|
146
|
-
size_t removed_index = (removed_entry - map->table);
|
147
|
-
struct hashmap_entry *entry;
|
148
|
-
|
149
|
-
/* Free the key */
|
150
|
-
if (map->key_free) {
|
151
|
-
map->key_free(removed_entry->key);
|
152
|
-
}
|
153
|
-
--map->num_entries;
|
154
|
-
|
155
|
-
/* Fill the free slot in the chain */
|
156
|
-
index = HASHMAP_PROBE_NEXT(map, removed_index);
|
157
|
-
for (i = 1; i < map->table_size; ++i) {
|
158
|
-
entry = &map->table[index];
|
159
|
-
if (!entry->key) {
|
160
|
-
/* Reached end of chain */
|
161
|
-
break;
|
162
|
-
}
|
163
|
-
entry_index = hashmap_calc_index(map, entry->key);
|
164
|
-
/* Shift in entries with an index <= to the removed slot */
|
165
|
-
if (HASHMAP_INDEX_LE(map, removed_index, entry_index)) {
|
166
|
-
#ifdef HASHMAP_METRICS
|
167
|
-
entry->num_collisions -= (i - removed_i);
|
168
|
-
removed_i = i;
|
169
|
-
#endif
|
170
|
-
memcpy(removed_entry, entry, sizeof(*removed_entry));
|
171
|
-
removed_index = index;
|
172
|
-
removed_entry = entry;
|
173
|
-
}
|
174
|
-
index = HASHMAP_PROBE_NEXT(map, index);
|
175
|
-
}
|
176
|
-
/* Clear the last removed entry */
|
177
|
-
memset(removed_entry, 0, sizeof(*removed_entry));
|
178
|
-
}
|
179
|
-
|
180
|
-
/*
|
181
|
-
* Reallocates the hash table to the new size and rehashes all entries.
|
182
|
-
* new_size MUST be a power of 2.
|
183
|
-
* Returns 0 on success and -errno on allocation or hash function failure.
|
184
|
-
*/
|
185
|
-
static int hashmap_rehash(struct hashmap *map, size_t new_size)
|
186
|
-
{
|
187
|
-
size_t old_size;
|
188
|
-
struct hashmap_entry *old_table;
|
189
|
-
struct hashmap_entry *new_table;
|
190
|
-
struct hashmap_entry *entry;
|
191
|
-
struct hashmap_entry *new_entry;
|
192
|
-
|
193
|
-
HASHMAP_ASSERT(new_size >= HASHMAP_SIZE_MIN);
|
194
|
-
HASHMAP_ASSERT((new_size & (new_size - 1)) == 0);
|
195
|
-
|
196
|
-
new_table = (struct hashmap_entry *)calloc(new_size,
|
197
|
-
sizeof(struct hashmap_entry));
|
198
|
-
if (!new_table) {
|
199
|
-
return -ENOMEM;
|
200
|
-
}
|
201
|
-
/* Backup old elements in case of rehash failure */
|
202
|
-
old_size = map->table_size;
|
203
|
-
old_table = map->table;
|
204
|
-
map->table_size = new_size;
|
205
|
-
map->table = new_table;
|
206
|
-
/* Rehash */
|
207
|
-
for (entry = old_table; entry < &old_table[old_size]; ++entry) {
|
208
|
-
if (!entry->data) {
|
209
|
-
/* Only copy entries with data */
|
210
|
-
continue;
|
211
|
-
}
|
212
|
-
new_entry = hashmap_entry_find(map, entry->key, true);
|
213
|
-
if (!new_entry) {
|
214
|
-
/*
|
215
|
-
* The load factor is too high with the new table
|
216
|
-
* size, or a poor hash function was used.
|
217
|
-
*/
|
218
|
-
goto revert;
|
219
|
-
}
|
220
|
-
/* Shallow copy (intentionally omits num_collisions) */
|
221
|
-
new_entry->key = entry->key;
|
222
|
-
new_entry->data = entry->data;
|
223
|
-
}
|
224
|
-
free(old_table);
|
225
|
-
return 0;
|
226
|
-
revert:
|
227
|
-
map->table_size = old_size;
|
228
|
-
map->table = old_table;
|
229
|
-
free(new_table);
|
230
|
-
return -EINVAL;
|
231
|
-
}
|
232
|
-
|
233
|
-
/*
|
234
|
-
* Iterate through all entries and free all keys.
|
235
|
-
*/
|
236
|
-
static void hashmap_free_keys(struct hashmap *map)
|
237
|
-
{
|
238
|
-
struct hashmap_iter *iter;
|
239
|
-
|
240
|
-
if (!map->key_free) {
|
241
|
-
return;
|
242
|
-
}
|
243
|
-
for (iter = hashmap_iter(map); iter;
|
244
|
-
iter = hashmap_iter_next(map, iter)) {
|
245
|
-
map->key_free((void *)hashmap_iter_get_key(iter));
|
246
|
-
}
|
247
|
-
}
|
248
|
-
|
249
|
-
/*
|
250
|
-
* Initialize an empty hashmap. A hash function and a key comparator are
|
251
|
-
* required.
|
252
|
-
*
|
253
|
-
* hash_func should return an even distribution of numbers between 0
|
254
|
-
* and SIZE_MAX varying on the key provided.
|
255
|
-
*
|
256
|
-
* key_compare_func should return 0 if the keys match, and non-zero otherwise.
|
257
|
-
*
|
258
|
-
* initial_size is optional, and may be set to the max number of entries
|
259
|
-
* expected to be put in the hash table. This is used as a hint to
|
260
|
-
* pre-allocate the hash table to the minimum size needed to avoid
|
261
|
-
* gratuitous rehashes. If initial_size 0, a default size will be used.
|
262
|
-
*
|
263
|
-
* Returns 0 on success and -errno on failure.
|
264
|
-
*/
|
265
|
-
int hashmap_init(struct hashmap *map, size_t (*hash_func)(const void *),
|
266
|
-
int (*key_compare_func)(const void *, const void *),
|
267
|
-
size_t initial_size)
|
268
|
-
{
|
269
|
-
HASHMAP_ASSERT(map != NULL);
|
270
|
-
HASHMAP_ASSERT(hash_func != NULL);
|
271
|
-
HASHMAP_ASSERT(key_compare_func != NULL);
|
272
|
-
|
273
|
-
if (!initial_size) {
|
274
|
-
initial_size = HASHMAP_SIZE_DEFAULT;
|
275
|
-
} else {
|
276
|
-
/* Convert init size to valid table size */
|
277
|
-
initial_size = hashmap_table_size_calc(initial_size);
|
278
|
-
}
|
279
|
-
map->table_size_init = initial_size;
|
280
|
-
map->table_size = initial_size;
|
281
|
-
map->num_entries = 0;
|
282
|
-
map->table = (struct hashmap_entry *)calloc(initial_size,
|
283
|
-
sizeof(struct hashmap_entry));
|
284
|
-
if (!map->table) {
|
285
|
-
return -ENOMEM;
|
286
|
-
}
|
287
|
-
map->hash = hash_func;
|
288
|
-
map->key_compare = key_compare_func;
|
289
|
-
map->key_alloc = NULL;
|
290
|
-
map->key_free = NULL;
|
291
|
-
return 0;
|
292
|
-
}
|
293
|
-
|
294
|
-
/*
|
295
|
-
* Free the hashmap and all associated memory.
|
296
|
-
*/
|
297
|
-
void hashmap_destroy(struct hashmap *map)
|
298
|
-
{
|
299
|
-
if (!map) {
|
300
|
-
return;
|
301
|
-
}
|
302
|
-
hashmap_free_keys(map);
|
303
|
-
free(map->table);
|
304
|
-
memset(map, 0, sizeof(*map));
|
305
|
-
}
|
306
|
-
|
307
|
-
/*
|
308
|
-
* Enable internal memory management of hash keys.
|
309
|
-
*/
|
310
|
-
void hashmap_set_key_alloc_funcs(struct hashmap *map,
|
311
|
-
void *(*key_alloc_func)(const void *),
|
312
|
-
void (*key_free_func)(void *))
|
313
|
-
{
|
314
|
-
HASHMAP_ASSERT(map != NULL);
|
315
|
-
|
316
|
-
map->key_alloc = key_alloc_func;
|
317
|
-
map->key_free = key_free_func;
|
318
|
-
}
|
319
|
-
|
320
|
-
/*
|
321
|
-
* Add an entry to the hashmap. If an entry with a matching key already
|
322
|
-
* exists and has a data pointer associated with it, the existing data
|
323
|
-
* pointer is returned, instead of assigning the new value. Compare
|
324
|
-
* the return value with the data passed in to determine if a new entry was
|
325
|
-
* created. Returns NULL if memory allocation failed.
|
326
|
-
*/
|
327
|
-
void *hashmap_put(struct hashmap *map, const void *key, void *data)
|
328
|
-
{
|
329
|
-
struct hashmap_entry *entry;
|
330
|
-
|
331
|
-
HASHMAP_ASSERT(map != NULL);
|
332
|
-
HASHMAP_ASSERT(key != NULL);
|
333
|
-
|
334
|
-
/* Rehash with 2x capacity if load factor is approaching 0.75 */
|
335
|
-
if (map->table_size <= hashmap_table_min_size_calc(map->num_entries)) {
|
336
|
-
hashmap_rehash(map, map->table_size << 1);
|
337
|
-
}
|
338
|
-
entry = hashmap_entry_find(map, key, true);
|
339
|
-
if (!entry) {
|
340
|
-
/*
|
341
|
-
* Cannot find an empty slot. Either out of memory, or using
|
342
|
-
* a poor hash function. Attempt to rehash once to reduce
|
343
|
-
* chain length.
|
344
|
-
*/
|
345
|
-
if (hashmap_rehash(map, map->table_size << 1) < 0) {
|
346
|
-
return NULL;
|
347
|
-
}
|
348
|
-
entry = hashmap_entry_find(map, key, true);
|
349
|
-
if (!entry) {
|
350
|
-
return NULL;
|
351
|
-
}
|
352
|
-
}
|
353
|
-
if (!entry->key) {
|
354
|
-
/* Allocate copy of key to simplify memory management */
|
355
|
-
if (map->key_alloc) {
|
356
|
-
entry->key = map->key_alloc(key);
|
357
|
-
if (!entry->key) {
|
358
|
-
return NULL;
|
359
|
-
}
|
360
|
-
} else {
|
361
|
-
entry->key = (void *)key;
|
362
|
-
}
|
363
|
-
++map->num_entries;
|
364
|
-
} else if (entry->data) {
|
365
|
-
/* Do not overwrite existing data */
|
366
|
-
return entry->data;
|
367
|
-
}
|
368
|
-
entry->data = data;
|
369
|
-
return data;
|
370
|
-
}
|
371
|
-
|
372
|
-
/*
|
373
|
-
* Return the data pointer, or NULL if no entry exists.
|
374
|
-
*/
|
375
|
-
void *hashmap_get(const struct hashmap *map, const void *key)
|
376
|
-
{
|
377
|
-
struct hashmap_entry *entry;
|
378
|
-
|
379
|
-
HASHMAP_ASSERT(map != NULL);
|
380
|
-
HASHMAP_ASSERT(key != NULL);
|
381
|
-
|
382
|
-
entry = hashmap_entry_find(map, key, false);
|
383
|
-
if (!entry) {
|
384
|
-
return NULL;
|
385
|
-
}
|
386
|
-
return entry->data;
|
387
|
-
}
|
388
|
-
|
389
|
-
/*
|
390
|
-
* Remove an entry with the specified key from the map.
|
391
|
-
* Returns the data pointer, or NULL, if no entry was found.
|
392
|
-
*/
|
393
|
-
void *hashmap_remove(struct hashmap *map, const void *key)
|
394
|
-
{
|
395
|
-
struct hashmap_entry *entry;
|
396
|
-
void *data;
|
397
|
-
|
398
|
-
HASHMAP_ASSERT(map != NULL);
|
399
|
-
HASHMAP_ASSERT(key != NULL);
|
400
|
-
|
401
|
-
entry = hashmap_entry_find(map, key, false);
|
402
|
-
if (!entry) {
|
403
|
-
return NULL;
|
404
|
-
}
|
405
|
-
data = entry->data;
|
406
|
-
/* Clear the entry and make the chain contiguous */
|
407
|
-
hashmap_entry_remove(map, entry);
|
408
|
-
return data;
|
409
|
-
}
|
410
|
-
|
411
|
-
/*
|
412
|
-
* Remove all entries.
|
413
|
-
*/
|
414
|
-
void hashmap_clear(struct hashmap *map)
|
415
|
-
{
|
416
|
-
HASHMAP_ASSERT(map != NULL);
|
417
|
-
|
418
|
-
hashmap_free_keys(map);
|
419
|
-
map->num_entries = 0;
|
420
|
-
memset(map->table, 0, sizeof(struct hashmap_entry) * map->table_size);
|
421
|
-
}
|
422
|
-
|
423
|
-
/*
|
424
|
-
* Remove all entries and reset the hash table to its initial size.
|
425
|
-
*/
|
426
|
-
void hashmap_reset(struct hashmap *map)
|
427
|
-
{
|
428
|
-
struct hashmap_entry *new_table;
|
429
|
-
|
430
|
-
HASHMAP_ASSERT(map != NULL);
|
431
|
-
|
432
|
-
hashmap_clear(map);
|
433
|
-
if (map->table_size == map->table_size_init) {
|
434
|
-
return;
|
435
|
-
}
|
436
|
-
new_table = (struct hashmap_entry *)realloc(map->table,
|
437
|
-
sizeof(struct hashmap_entry) * map->table_size_init);
|
438
|
-
if (!new_table) {
|
439
|
-
return;
|
440
|
-
}
|
441
|
-
map->table = new_table;
|
442
|
-
map->table_size = map->table_size_init;
|
443
|
-
}
|
444
|
-
|
445
|
-
/*
|
446
|
-
* Return the number of entries in the hash map.
|
447
|
-
*/
|
448
|
-
size_t hashmap_size(const struct hashmap *map)
|
449
|
-
{
|
450
|
-
HASHMAP_ASSERT(map != NULL);
|
451
|
-
|
452
|
-
return map->num_entries;
|
453
|
-
}
|
454
|
-
|
455
|
-
/*
|
456
|
-
* Get a new hashmap iterator. The iterator is an opaque
|
457
|
-
* pointer that may be used with hashmap_iter_*() functions.
|
458
|
-
* Hashmap iterators are INVALID after a put or remove operation is performed.
|
459
|
-
* hashmap_iter_remove() allows safe removal during iteration.
|
460
|
-
*/
|
461
|
-
struct hashmap_iter *hashmap_iter(const struct hashmap *map)
|
462
|
-
{
|
463
|
-
HASHMAP_ASSERT(map != NULL);
|
464
|
-
|
465
|
-
if (!map->num_entries) {
|
466
|
-
return NULL;
|
467
|
-
}
|
468
|
-
return (struct hashmap_iter *)hashmap_entry_get_populated(map,
|
469
|
-
map->table);
|
470
|
-
}
|
471
|
-
|
472
|
-
/*
|
473
|
-
* Return an iterator to the next hashmap entry. Returns NULL if there are
|
474
|
-
* no more entries.
|
475
|
-
*/
|
476
|
-
struct hashmap_iter *hashmap_iter_next(const struct hashmap *map,
|
477
|
-
const struct hashmap_iter *iter)
|
478
|
-
{
|
479
|
-
struct hashmap_entry *entry = (struct hashmap_entry *)iter;
|
480
|
-
|
481
|
-
HASHMAP_ASSERT(map != NULL);
|
482
|
-
|
483
|
-
if (!iter) {
|
484
|
-
return NULL;
|
485
|
-
}
|
486
|
-
return (struct hashmap_iter *)hashmap_entry_get_populated(map,
|
487
|
-
entry + 1);
|
488
|
-
}
|
489
|
-
|
490
|
-
/*
|
491
|
-
* Remove the hashmap entry pointed to by this iterator and return an
|
492
|
-
* iterator to the next entry. Returns NULL if there are no more entries.
|
493
|
-
*/
|
494
|
-
struct hashmap_iter *hashmap_iter_remove(struct hashmap *map,
|
495
|
-
const struct hashmap_iter *iter)
|
496
|
-
{
|
497
|
-
struct hashmap_entry *entry = (struct hashmap_entry *)iter;
|
498
|
-
|
499
|
-
HASHMAP_ASSERT(map != NULL);
|
500
|
-
|
501
|
-
if (!iter) {
|
502
|
-
return NULL;
|
503
|
-
}
|
504
|
-
if (!entry->key) {
|
505
|
-
/* Iterator is invalid, so just return the next valid entry */
|
506
|
-
return hashmap_iter_next(map, iter);
|
507
|
-
}
|
508
|
-
hashmap_entry_remove(map, entry);
|
509
|
-
return (struct hashmap_iter *)hashmap_entry_get_populated(map, entry);
|
510
|
-
}
|
511
|
-
|
512
|
-
/*
|
513
|
-
* Return the key of the entry pointed to by the iterator.
|
514
|
-
*/
|
515
|
-
const void *hashmap_iter_get_key(const struct hashmap_iter *iter)
|
516
|
-
{
|
517
|
-
if (!iter) {
|
518
|
-
return NULL;
|
519
|
-
}
|
520
|
-
return (const void *)((struct hashmap_entry *)iter)->key;
|
521
|
-
}
|
522
|
-
|
523
|
-
/*
|
524
|
-
* Return the data of the entry pointed to by the iterator.
|
525
|
-
*/
|
526
|
-
void *hashmap_iter_get_data(const struct hashmap_iter *iter)
|
527
|
-
{
|
528
|
-
if (!iter) {
|
529
|
-
return NULL;
|
530
|
-
}
|
531
|
-
return ((struct hashmap_entry *)iter)->data;
|
532
|
-
}
|
533
|
-
|
534
|
-
/*
|
535
|
-
* Set the data pointer of the entry pointed to by the iterator.
|
536
|
-
*/
|
537
|
-
void hashmap_iter_set_data(const struct hashmap_iter *iter, void *data)
|
538
|
-
{
|
539
|
-
if (!iter) {
|
540
|
-
return;
|
541
|
-
}
|
542
|
-
((struct hashmap_entry *)iter)->data = data;
|
543
|
-
}
|
544
|
-
|
545
|
-
/*
|
546
|
-
* Invoke func for each entry in the hashmap. Unlike the hashmap_iter_*()
|
547
|
-
* interface, this function supports calls to hashmap_remove() during iteration.
|
548
|
-
* However, it is an error to put or remove an entry other than the current one,
|
549
|
-
* and doing so will immediately halt iteration and return an error.
|
550
|
-
* Iteration is stopped if func returns non-zero. Returns func's return
|
551
|
-
* value if it is < 0, otherwise, 0.
|
552
|
-
*/
|
553
|
-
int hashmap_foreach(const struct hashmap *map,
|
554
|
-
int (*func)(const void *, void *, void *), void *arg)
|
555
|
-
{
|
556
|
-
struct hashmap_entry *entry;
|
557
|
-
size_t num_entries;
|
558
|
-
const void *key;
|
559
|
-
int rc;
|
560
|
-
|
561
|
-
HASHMAP_ASSERT(map != NULL);
|
562
|
-
HASHMAP_ASSERT(func != NULL);
|
563
|
-
|
564
|
-
entry = map->table;
|
565
|
-
for (entry = map->table; entry < &map->table[map->table_size];
|
566
|
-
++entry) {
|
567
|
-
if (!entry->key) {
|
568
|
-
continue;
|
569
|
-
}
|
570
|
-
num_entries = map->num_entries;
|
571
|
-
key = entry->key;
|
572
|
-
rc = func(entry->key, entry->data, arg);
|
573
|
-
if (rc < 0) {
|
574
|
-
return rc;
|
575
|
-
}
|
576
|
-
if (rc > 0) {
|
577
|
-
return 0;
|
578
|
-
}
|
579
|
-
/* Run this entry again if func() deleted it */
|
580
|
-
if (entry->key != key) {
|
581
|
-
--entry;
|
582
|
-
} else if (num_entries != map->num_entries) {
|
583
|
-
/* Stop immediately if func put/removed another entry */
|
584
|
-
return -1;
|
585
|
-
}
|
586
|
-
}
|
587
|
-
return 0;
|
588
|
-
}
|
589
|
-
|
590
|
-
/*
|
591
|
-
* Default hash function for string keys.
|
592
|
-
* This is an implementation of the well-documented Jenkins one-at-a-time
|
593
|
-
* hash function.
|
594
|
-
*/
|
595
|
-
size_t hashmap_hash_string(const void *key)
|
596
|
-
{
|
597
|
-
const char *key_str = (const char *)key;
|
598
|
-
size_t hash = 0;
|
599
|
-
|
600
|
-
for (; *key_str; ++key_str) {
|
601
|
-
hash += *key_str;
|
602
|
-
hash += (hash << 10);
|
603
|
-
hash ^= (hash >> 6);
|
604
|
-
}
|
605
|
-
hash += (hash << 3);
|
606
|
-
hash ^= (hash >> 11);
|
607
|
-
hash += (hash << 15);
|
608
|
-
return hash;
|
609
|
-
}
|
610
|
-
|
611
|
-
/*
|
612
|
-
* Default key comparator function for string keys.
|
613
|
-
*/
|
614
|
-
int hashmap_compare_string(const void *a, const void *b)
|
615
|
-
{
|
616
|
-
return strcmp((const char *)a, (const char *)b);
|
617
|
-
}
|
618
|
-
|
619
|
-
/*
|
620
|
-
* Default key allocation function for string keys. Use free() for the
|
621
|
-
* key_free_func.
|
622
|
-
*/
|
623
|
-
void *hashmap_alloc_key_string(const void *key)
|
624
|
-
{
|
625
|
-
return (void *)strdup((const char *)key);
|
626
|
-
}
|
627
|
-
|
628
|
-
#ifdef HASHMAP_METRICS
|
629
|
-
/*
|
630
|
-
* Return the load factor.
|
631
|
-
*/
|
632
|
-
double hashmap_load_factor(const struct hashmap *map)
|
633
|
-
{
|
634
|
-
HASHMAP_ASSERT(map != NULL);
|
635
|
-
|
636
|
-
if (!map->table_size) {
|
637
|
-
return 0;
|
638
|
-
}
|
639
|
-
return (double)map->num_entries / map->table_size;
|
640
|
-
}
|
641
|
-
|
642
|
-
/*
|
643
|
-
* Return the average number of collisions per entry.
|
644
|
-
*/
|
645
|
-
double hashmap_collisions_mean(const struct hashmap *map)
|
646
|
-
{
|
647
|
-
struct hashmap_entry *entry;
|
648
|
-
size_t total_collisions = 0;
|
649
|
-
|
650
|
-
HASHMAP_ASSERT(map != NULL);
|
651
|
-
|
652
|
-
if (!map->num_entries) {
|
653
|
-
return 0;
|
654
|
-
}
|
655
|
-
for (entry = map->table; entry < &map->table[map->table_size];
|
656
|
-
++entry) {
|
657
|
-
if (!entry->key) {
|
658
|
-
continue;
|
659
|
-
}
|
660
|
-
total_collisions += entry->num_collisions;
|
661
|
-
}
|
662
|
-
return (double)total_collisions / map->num_entries;
|
663
|
-
}
|
664
|
-
|
665
|
-
/*
|
666
|
-
* Return the variance between entry collisions. The higher the variance,
|
667
|
-
* the more likely the hash function is poor and is resulting in clustering.
|
668
|
-
*/
|
669
|
-
double hashmap_collisions_variance(const struct hashmap *map)
|
670
|
-
{
|
671
|
-
struct hashmap_entry *entry;
|
672
|
-
double mean_collisions;
|
673
|
-
double variance;
|
674
|
-
double total_variance = 0;
|
675
|
-
|
676
|
-
HASHMAP_ASSERT(map != NULL);
|
677
|
-
|
678
|
-
if (!map->num_entries) {
|
679
|
-
return 0;
|
680
|
-
}
|
681
|
-
mean_collisions = hashmap_collisions_mean(map);
|
682
|
-
for (entry = map->table; entry < &map->table[map->table_size];
|
683
|
-
++entry) {
|
684
|
-
if (!entry->key) {
|
685
|
-
continue;
|
686
|
-
}
|
687
|
-
variance = (double)entry->num_collisions - mean_collisions;
|
688
|
-
total_variance += variance * variance;
|
689
|
-
}
|
690
|
-
return total_variance / map->num_entries;
|
691
|
-
}
|
692
|
-
#endif
|