1Gc(3) OCaml library Gc(3)
2
6 Gc - Memory management control and statistics; finalised values.
7
9 Module Gc
10
12 Module Gc
13 : sig end
14 Memory management control and statistics; finalised values.
17 type stat = {
23 minor_words : float ; (* Number of words allocated in the minor heap
24 since the program was started.
25 *)
26 promoted_words : float ; (* Number of words allocated in the minor
27 heap that survived a minor collection and were moved to the major heap
28 since the program was started.
29 *)
30 major_words : float ; (* Number of words allocated in the major heap,
31 including the promoted words, since the program was started.
32 *)
33 minor_collections : int ; (* Number of minor collections since the
34 program was started.
35 *)
36 major_collections : int ; (* Number of major collection cycles com‐
37 pleted since the program was started.
38 *)
39 heap_words : int ; (* Total size of the major heap, in words.
40 *)
41 heap_chunks : int ; (* Number of contiguous pieces of memory that
42 make up the major heap.
43 *)
44 live_words : int ; (* Number of words of live data in the major heap,
45 including the header words.
46 Note that "live" words refers to every word in the major heap that
48 isn't currently known to be collectable, which includes words that have
49 become unreachable by the program after the start of the previous gc
50 cycle. It is typically much simpler and more predictable to call
51 Gc.full_major (or Gc.compact ) then computing gc stats, as then "live"
52 words has the simple meaning of "reachable by the program". One caveat
53 is that a single call to Gc.full_major will not reclaim values that
54 have a finaliser from Gc.finalise (this does not apply to Gc.fi‐
55 nalise_last ). If this caveat matters, simply call Gc.full_major twice
56 instead of once.
57 *)
58 live_blocks : int ; (* Number of live blocks in the major heap.
59 See live_words for a caveat about what "live" means.
61 *)
62 free_words : int ; (* Number of words in the free list.
63 *)
64 free_blocks : int ; (* Number of blocks in the free list.
65 *)
66 largest_free : int ; (* Size (in words) of the largest block in the
67 free list.
68 *)
69 fragments : int ; (* Number of wasted words due to fragmentation.
70 These are 1-words free blocks placed between two live blocks. They are
71 not available for allocation.
72 *)
73 compactions : int ; (* Number of heap compactions since the program
74 was started.
75 *)
76 top_heap_words : int ; (* Maximum size reached by the major heap, in
77 words.
78 *)
79 stack_size : int ; (* Current size of the stack, in words.
80 Since 3.12.0
83 *)
84 forced_major_collections : int ; (* Number of forced full major col‐
85 lections completed since the program was started.
86 Since 4.12.0
89 *)
90 }
91 The memory management counters are returned in a stat record. These
94 counters give values for the whole program.
95 The total amount of memory allocated by the program since it was
97 started is (in words) minor_words + major_words - promoted_words .
98 Multiply by the word size (4 on a 32-bit machine, 8 on a 64-bit ma‐
99 chine) to get the number of bytes.
100 type control = {
103 minor_heap_size : int ; (* The size (in words) of the minor heap.
104 Changing this parameter will trigger a minor collection. The total size
105 of the minor heap used by this program is the sum of the heap sizes of
106 the active domains. Default: 256k.
107 *)
108 major_heap_increment : int ; (* How much to add to the major heap
109 when increasing it. If this number is less than or equal to 1000, it is
110 a percentage of the current heap size (i.e. setting it to 100 will dou‐
111 ble the heap size at each increase). If it is more than 1000, it is a
112 fixed number of words that will be added to the heap. Default: 15.
113 *)
114 space_overhead : int ; (* The major GC speed is computed from this
115 parameter. This is the memory that will be "wasted" because the GC
116 does not immediately collect unreachable blocks. It is expressed as a
117 percentage of the memory used for live data. The GC will work more
118 (use more CPU time and collect blocks more eagerly) if space_overhead
119 is smaller. Default: 120.
120 *)
121 verbose : int ; (* This value controls the GC messages on standard
122 error output. It is a sum of some of the following flags, to print
123 messages on the corresponding events:
124 - 0x001 Start and end of major GC cycle.
126 - 0x002 Minor collection and major GC slice.
128 - 0x004 Growing and shrinking of the heap.
130 - 0x008 Resizing of stacks and memory manager tables.
132 - 0x010 Heap compaction.
134 - 0x020 Change of GC parameters.
136 - 0x040 Computation of major GC slice size.
138 - 0x080 Calling of finalisation functions.
140 - 0x100 Bytecode executable and shared library search at start-up.
142 - 0x200 Computation of compaction-triggering condition.
144 - 0x400 Output GC statistics at program exit. Default: 0.
146 *)
148 max_overhead : int ; (* Heap compaction is triggered when the esti‐
149 mated amount of "wasted" memory is more than max_overhead percent of
150 the amount of live data. If max_overhead is set to 0, heap compaction
151 is triggered at the end of each major GC cycle (this setting is in‐
152 tended for testing purposes only). If max_overhead >= 1000000 , com‐
153 paction is never triggered. If compaction is permanently disabled, it
154 is strongly suggested to set allocation_policy to 2. Default: 500.
155 *)
156 stack_limit : int ; (* The maximum size of the fiber stacks (in
157 words). Default: 1024k.
158 *)
159 allocation_policy : int ; (* The policy used for allocating in the
160 major heap. Possible values are 0, 1 and 2.
161 -0 is the next-fit policy, which is usually fast but can result in
164 fragmentation, increasing memory consumption.
165 -1 is the first-fit policy, which avoids fragmentation but has corner
168 cases (in certain realistic workloads) where it is sensibly slower.
169 -2 is the best-fit policy, which is fast and avoids fragmentation. In
172 our experiments it is faster and uses less memory than both next-fit
173 and first-fit. (since OCaml 4.10)
174 The default is best-fit.
176 On one example that was known to be bad for next-fit and first-fit,
178 next-fit takes 28s using 855Mio of memory, first-fit takes 47s using
179 566Mio of memory, best-fit takes 27s using 545Mio of memory.
180 Note: If you change to next-fit, you may need to reduce the space_over‐
182 head setting, for example using 80 instead of the default 120 which is
183 tuned for best-fit. Otherwise, your program will need more memory.
184 Note: changing the allocation policy at run-time forces a heap com‐
186 paction, which is a lengthy operation unless the heap is small (e.g. at
187 the start of the program).
188 Default: 2.
190 Since 3.11.0
193 *)
194 window_size : int ; (* The size of the window used by the major GC
195 for smoothing out variations in its workload. This is an integer be‐
196 tween 1 and 50. Default: 1.
197 Since 4.03.0
200 *)
201 custom_major_ratio : int ; (* Target ratio of floating garbage to ma‐
202 jor heap size for out-of-heap memory held by custom values located in
203 the major heap. The GC speed is adjusted to try to use this much memory
204 for dead values that are not yet collected. Expressed as a percentage
205 of major heap size. The default value keeps the out-of-heap floating
206 garbage about the same size as the in-heap overhead. Note: this only
207 applies to values allocated with caml_alloc_custom_mem (e.g. bigar‐
208 rays). Default: 44.
209 Since 4.08.0
212 *)
213 custom_minor_ratio : int ; (* Bound on floating garbage for
214 out-of-heap memory held by custom values in the minor heap. A minor GC
215 is triggered when this much memory is held by custom values located in
216 the minor heap. Expressed as a percentage of minor heap size. Note:
217 this only applies to values allocated with caml_alloc_custom_mem (e.g.
218 bigarrays). Default: 100.
219 Since 4.08.0
222 *)
223 custom_minor_max_size : int ; (* Maximum amount of out-of-heap memory
224 for each custom value allocated in the minor heap. When a custom value
225 is allocated on the minor heap and holds more than this many bytes,
226 only this value is counted against custom_minor_ratio and the rest is
227 directly counted against custom_major_ratio . Note: this only applies
228 to values allocated with caml_alloc_custom_mem (e.g. bigarrays). De‐
229 fault: 8192 bytes.
230 Since 4.08.0
233 *)
234 }
235 The GC parameters are given as a control record. Note that these pa‐
238 rameters can also be initialised by setting the OCAMLRUNPARAM environ‐
239 ment variable. See the documentation of ocamlrun .
240 val stat : unit -> stat
244 Return the current values of the memory management counters in a stat
246 record that represent the program's total memory stats. This function
247 causes a full major collection.
248 val quick_stat : unit -> stat
252 Same as stat except that live_words , live_blocks , free_words ,
254 free_blocks , largest_free , and fragments are set to 0. Due to per-do‐
255 main buffers it may only represent the state of the program's total
256 memory usage since the last minor collection. This function is much
257 faster than stat because it does not need to trigger a full major col‐
258 lection.
259 val counters : unit -> float * float * float
263 Return (minor_words, promoted_words, major_words) for the current do‐
265 main or potentially previous domains. This function is as fast as
266 quick_stat .
267 val minor_words : unit -> float
271 Number of words allocated in the minor heap by this domain or poten‐
273 tially previous domains. This number is accurate in byte-code programs,
274 but only an approximation in programs compiled to native code.
275 In native code this function does not allocate.
277 Since 4.04
280 val get : unit -> control
284 Return the current values of the GC parameters in a control record.
286 Alert unsynchronized_access. GC parameters are a mutable global state.
289 val set : control -> unit
293 set r changes the GC parameters according to the control record r .
296 The normal usage is: Gc.set { (Gc.get()) with Gc.verbose = 0x00d }
297 Alert unsynchronized_access. GC parameters are a mutable global state.
301 val minor : unit -> unit
305 Trigger a minor collection.
307 val major_slice : int -> int
311 major_slice n Do a minor collection and a slice of major collection. n
314 is the size of the slice: the GC will do enough work to free (on aver‐
315 age) n words of memory. If n = 0, the GC will try to do enough work to
316 ensure that the next automatic slice has no work to do. This function
317 returns an unspecified integer (currently: 0).
318 val major : unit -> unit
322 Do a minor collection and finish the current major collection cycle.
324 val full_major : unit -> unit
328 Do a minor collection, finish the current major collection cycle, and
330 perform a complete new cycle. This will collect all currently unreach‐
331 able blocks.
332 val compact : unit -> unit
336 Perform a full major collection and compact the heap. Note that heap
338 compaction is a lengthy operation.
339 val print_stat : out_channel -> unit
343 Print the current values of the memory management counters (in hu‐
345 man-readable form) of the total program into the channel argument.
346 val allocated_bytes : unit -> float
350 Return the number of bytes allocated by this domain and potentially a
352 previous domain. It is returned as a float to avoid overflow problems
353 with int on 32-bit machines.
354 val get_minor_free : unit -> int
358 Return the current size of the free space inside the minor heap of this
360 domain.
361 Since 4.03.0
364 val finalise : ('a -> unit) -> 'a -> unit
368 finalise f v registers f as a finalisation function for v . v must be
371 heap-allocated. f will be called with v as argument at some point be‐
372 tween the first time v becomes unreachable (including through weak
373 pointers) and the time v is collected by the GC. Several functions can
374 be registered for the same value, or even several instances of the same
375 function. Each instance will be called once (or never, if the program
376 terminates before v becomes unreachable).
377 The GC will call the finalisation functions in the order of dealloca‐
379 tion. When several values become unreachable at the same time (i.e.
380 during the same GC cycle), the finalisation functions will be called in
381 the reverse order of the corresponding calls to finalise . If finalise
382 is called in the same order as the values are allocated, that means
383 each value is finalised before the values it depends upon. Of course,
384 this becomes false if additional dependencies are introduced by assign‐
385 ments.
386 In the presence of multiple OCaml threads it should be assumed that any
388 particular finaliser may be executed in any of the threads.
389 Anything reachable from the closure of finalisation functions is con‐
391 sidered reachable, so the following code will not work as expected:
392 - let v = ... in Gc.finalise (fun _ -> ...v...) v
394 Instead you should make sure that v is not in the closure of the final‐
396 isation function by writing:
397 - let f = fun x -> ... let v = ... in Gc.finalise f v
399 The f function can use all features of OCaml, including assignments
401 that make the value reachable again. It can also loop forever (in this
402 case, the other finalisation functions will not be called during the
403 execution of f, unless it calls finalise_release ). It can call fi‐
404 nalise on v or other values to register other functions or even itself.
405 It can raise an exception; in this case the exception will interrupt
406 whatever the program was doing when the function was called.
407 finalise will raise Invalid_argument if v is not guaranteed to be
410 heap-allocated. Some examples of values that are not heap-allocated
411 are integers, constant constructors, booleans, the empty array, the
412 empty list, the unit value. The exact list of what is heap-allocated
413 or not is implementation-dependent. Some constant values can be
414 heap-allocated but never deallocated during the lifetime of the pro‐
415 gram, for example a list of integer constants; this is also implementa‐
416 tion-dependent. Note that values of types float are sometimes allo‐
417 cated and sometimes not, so finalising them is unsafe, and finalise
418 will also raise Invalid_argument for them. Values of type 'a Lazy.t
419 (for any 'a ) are like float in this respect, except that the compiler
420 sometimes optimizes them in a way that prevents finalise from detecting
421 them. In this case, it will not raise Invalid_argument , but you should
422 still avoid calling finalise on lazy values.
423 The results of calling String.make , Bytes.make , Bytes.create , Ar‐
425 ray.make , and ref are guaranteed to be heap-allocated and non-constant
426 except when the length argument is 0 .
427 val finalise_last : (unit -> unit) -> 'a -> unit
431 same as Gc.finalise except the value is not given as argument. So you
433 can't use the given value for the computation of the finalisation func‐
434 tion. The benefit is that the function is called after the value is un‐
435 reachable for the last time instead of the first time. So contrary to
436 Gc.finalise the value will never be reachable again or used again. In
437 particular every weak pointer and ephemeron that contained this value
438 as key or data is unset before running the finalisation function. More‐
439 over the finalisation functions attached with Gc.finalise are always
440 called before the finalisation functions attached with Gc.finalise_last
441 .
442 Since 4.04
445 val finalise_release : unit -> unit
449 A finalisation function may call finalise_release to tell the GC that
451 it can launch the next finalisation function without waiting for the
452 current one to return.
453 type alarm
456 An alarm is a piece of data that calls a user function at the end of
459 each major GC cycle. The following functions are provided to create
460 and delete alarms.
461 val create_alarm : (unit -> unit) -> alarm
465 create_alarm f will arrange for f to be called at the end of each major
468 GC cycle, starting with the current cycle or the next one. A value of
469 type alarm is returned that you can use to call delete_alarm .
470 val delete_alarm : alarm -> unit
474 delete_alarm a will stop the calls to the function associated to a .
477 Calling delete_alarm a again has no effect.
478 val eventlog_pause : unit -> unit
482 Deprecated. Use Runtime_events.pause instead.
484 val eventlog_resume : unit -> unit
488 Deprecated. Use Runtime_events.resume instead.
490 module Memprof : sig end
493 Memprof is a sampling engine for allocated memory words. Every allo‐
497 cated word has a probability of being sampled equal to a configurable
498 sampling rate. Once a block is sampled, it becomes tracked. A tracked
499 block triggers a user-defined callback as soon as it is allocated, pro‐
500 moted or deallocated.
501 Since blocks are composed of several words, a block can potentially be
503 sampled several times. If a block is sampled several times, then each
504 of the callback is called once for each event of this block: the multi‐
505 plicity is given in the n_samples field of the allocation structure.
506 This engine makes it possible to implement a low-overhead memory pro‐
508 filer as an OCaml library.
509 Note: this API is EXPERIMENTAL. It may change without prior notice.
511OCamldoc 2023-07-20 Gc(3)