Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_prewarm.c
4 : * prewarming utilities
5 : *
6 : * Copyright (c) 2010-2026, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * contrib/pg_prewarm/pg_prewarm.c
10 : *
11 : *-------------------------------------------------------------------------
12 : */
13 : #include "postgres.h"
14 :
15 : #include <sys/stat.h>
16 : #include <unistd.h>
17 :
18 : #include "access/relation.h"
19 : #include "catalog/index.h"
20 : #include "fmgr.h"
21 : #include "miscadmin.h"
22 : #include "storage/bufmgr.h"
23 : #include "storage/lmgr.h"
24 : #include "storage/read_stream.h"
25 : #include "storage/smgr.h"
26 : #include "utils/acl.h"
27 : #include "utils/builtins.h"
28 : #include "utils/lsyscache.h"
29 : #include "utils/rel.h"
30 :
31 0 : PG_MODULE_MAGIC_EXT(
32 : .name = "pg_prewarm",
33 : .version = PG_VERSION
34 : );
35 :
36 0 : PG_FUNCTION_INFO_V1(pg_prewarm);
37 :
38 : typedef enum
39 : {
40 : PREWARM_PREFETCH,
41 : PREWARM_READ,
42 : PREWARM_BUFFER,
43 : } PrewarmType;
44 :
45 : static PGIOAlignedBlock blockbuffer;
46 :
47 : /*
48 : * pg_prewarm(regclass, mode text, fork text,
49 : * first_block int8, last_block int8)
50 : *
51 : * The first argument is the relation to be prewarmed; the second controls
52 : * how prewarming is done; legal options are 'prefetch', 'read', and 'buffer'.
53 : * The third is the name of the relation fork to be prewarmed. The fourth
54 : * and fifth arguments specify the first and last block to be prewarmed.
55 : * If the fourth argument is NULL, it will be taken as 0; if the fifth argument
56 : * is NULL, it will be taken as the number of blocks in the relation. The
57 : * return value is the number of blocks successfully prewarmed.
58 : */
59 : Datum
60 0 : pg_prewarm(PG_FUNCTION_ARGS)
61 : {
62 0 : Oid relOid;
63 0 : text *forkName;
64 0 : text *type;
65 0 : int64 first_block;
66 0 : int64 last_block;
67 0 : int64 nblocks;
68 0 : int64 blocks_done = 0;
69 0 : int64 block;
70 0 : Relation rel;
71 0 : ForkNumber forkNumber;
72 0 : char *forkString;
73 0 : char *ttype;
74 0 : PrewarmType ptype;
75 0 : AclResult aclresult;
76 0 : char relkind;
77 0 : Oid privOid;
78 :
79 : /* Basic sanity checking. */
80 0 : if (PG_ARGISNULL(0))
81 0 : ereport(ERROR,
82 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
83 : errmsg("relation cannot be null")));
84 0 : relOid = PG_GETARG_OID(0);
85 0 : if (PG_ARGISNULL(1))
86 0 : ereport(ERROR,
87 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
88 : errmsg("prewarm type cannot be null")));
89 0 : type = PG_GETARG_TEXT_PP(1);
90 0 : ttype = text_to_cstring(type);
91 0 : if (strcmp(ttype, "prefetch") == 0)
92 0 : ptype = PREWARM_PREFETCH;
93 0 : else if (strcmp(ttype, "read") == 0)
94 0 : ptype = PREWARM_READ;
95 0 : else if (strcmp(ttype, "buffer") == 0)
96 0 : ptype = PREWARM_BUFFER;
97 : else
98 : {
99 0 : ereport(ERROR,
100 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
101 : errmsg("invalid prewarm type"),
102 : errhint("Valid prewarm types are \"prefetch\", \"read\", and \"buffer\".")));
103 0 : PG_RETURN_INT64(0); /* Placate compiler. */
104 : }
105 0 : if (PG_ARGISNULL(2))
106 0 : ereport(ERROR,
107 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
108 : errmsg("relation fork cannot be null")));
109 0 : forkName = PG_GETARG_TEXT_PP(2);
110 0 : forkString = text_to_cstring(forkName);
111 0 : forkNumber = forkname_to_number(forkString);
112 :
113 : /*
114 : * Open relation and check privileges. If the relation is an index, we
115 : * must check the privileges on its parent table instead.
116 : */
117 0 : relkind = get_rel_relkind(relOid);
118 0 : if (relkind == RELKIND_INDEX ||
119 0 : relkind == RELKIND_PARTITIONED_INDEX)
120 : {
121 0 : privOid = IndexGetRelation(relOid, true);
122 :
123 : /* Lock table before index to avoid deadlock. */
124 0 : if (OidIsValid(privOid))
125 0 : LockRelationOid(privOid, AccessShareLock);
126 0 : }
127 : else
128 0 : privOid = relOid;
129 :
130 0 : rel = relation_open(relOid, AccessShareLock);
131 :
132 : /*
133 : * It's possible that the relation with OID "privOid" was dropped and the
134 : * OID was reused before we locked it. If that happens, we could be left
135 : * with the wrong parent table OID, in which case we must ERROR. It's
136 : * possible that such a race would change the outcome of
137 : * get_rel_relkind(), too, but the worst case scenario there is that we'll
138 : * check privileges on the index instead of its parent table, which isn't
139 : * too terrible.
140 : */
141 0 : if (!OidIsValid(privOid) ||
142 0 : (privOid != relOid &&
143 0 : privOid != IndexGetRelation(relOid, true)))
144 0 : ereport(ERROR,
145 : (errcode(ERRCODE_UNDEFINED_TABLE),
146 : errmsg("could not find parent table of index \"%s\"",
147 : RelationGetRelationName(rel))));
148 :
149 0 : aclresult = pg_class_aclcheck(privOid, GetUserId(), ACL_SELECT);
150 0 : if (aclresult != ACLCHECK_OK)
151 0 : aclcheck_error(aclresult, get_relkind_objtype(rel->rd_rel->relkind), get_rel_name(relOid));
152 :
153 : /* Check that the relation has storage. */
154 0 : if (!RELKIND_HAS_STORAGE(rel->rd_rel->relkind))
155 0 : ereport(ERROR,
156 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
157 : errmsg("relation \"%s\" does not have storage",
158 : RelationGetRelationName(rel)),
159 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
160 :
161 : /* Check that the fork exists. */
162 0 : if (!smgrexists(RelationGetSmgr(rel), forkNumber))
163 0 : ereport(ERROR,
164 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
165 : errmsg("fork \"%s\" does not exist for this relation",
166 : forkString)));
167 :
168 : /* Validate block numbers, or handle nulls. */
169 0 : nblocks = RelationGetNumberOfBlocksInFork(rel, forkNumber);
170 0 : if (PG_ARGISNULL(3))
171 0 : first_block = 0;
172 : else
173 : {
174 0 : first_block = PG_GETARG_INT64(3);
175 0 : if (first_block < 0 || first_block >= nblocks)
176 0 : ereport(ERROR,
177 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
178 : errmsg("starting block number must be between 0 and %" PRId64,
179 : (nblocks - 1))));
180 : }
181 0 : if (PG_ARGISNULL(4))
182 0 : last_block = nblocks - 1;
183 : else
184 : {
185 0 : last_block = PG_GETARG_INT64(4);
186 0 : if (last_block < 0 || last_block >= nblocks)
187 0 : ereport(ERROR,
188 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
189 : errmsg("ending block number must be between 0 and %" PRId64,
190 : (nblocks - 1))));
191 : }
192 :
193 : /* Now we're ready to do the real work. */
194 0 : if (ptype == PREWARM_PREFETCH)
195 : {
196 : #ifdef USE_PREFETCH
197 :
198 : /*
199 : * In prefetch mode, we just hint the OS to read the blocks, but we
200 : * don't know whether it really does it, and we don't wait for it to
201 : * finish.
202 : *
203 : * It would probably be better to pass our prefetch requests in chunks
204 : * of a megabyte or maybe even a whole segment at a time, but there's
205 : * no practical way to do that at present without a gross modularity
206 : * violation, so we just do this.
207 : */
208 0 : for (block = first_block; block <= last_block; ++block)
209 : {
210 0 : CHECK_FOR_INTERRUPTS();
211 0 : PrefetchBuffer(rel, forkNumber, block);
212 0 : ++blocks_done;
213 0 : }
214 : #else
215 : ereport(ERROR,
216 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
217 : errmsg("prefetch is not supported by this build")));
218 : #endif
219 0 : }
220 0 : else if (ptype == PREWARM_READ)
221 : {
222 : /*
223 : * In read mode, we actually read the blocks, but not into shared
224 : * buffers. This is more portable than prefetch mode (it works
225 : * everywhere) and is synchronous.
226 : */
227 0 : for (block = first_block; block <= last_block; ++block)
228 : {
229 0 : CHECK_FOR_INTERRUPTS();
230 0 : smgrread(RelationGetSmgr(rel), forkNumber, block, blockbuffer.data);
231 0 : ++blocks_done;
232 0 : }
233 0 : }
234 0 : else if (ptype == PREWARM_BUFFER)
235 : {
236 0 : BlockRangeReadStreamPrivate p;
237 0 : ReadStream *stream;
238 :
239 : /*
240 : * In buffer mode, we actually pull the data into shared_buffers.
241 : */
242 :
243 : /* Set up the private state for our streaming buffer read callback. */
244 0 : p.current_blocknum = first_block;
245 0 : p.last_exclusive = last_block + 1;
246 :
247 : /*
248 : * It is safe to use batchmode as block_range_read_stream_cb takes no
249 : * locks.
250 : */
251 0 : stream = read_stream_begin_relation(READ_STREAM_MAINTENANCE |
252 : READ_STREAM_FULL |
253 : READ_STREAM_USE_BATCHING,
254 : NULL,
255 0 : rel,
256 0 : forkNumber,
257 : block_range_read_stream_cb,
258 : &p,
259 : 0);
260 :
261 0 : for (block = first_block; block <= last_block; ++block)
262 : {
263 0 : Buffer buf;
264 :
265 0 : CHECK_FOR_INTERRUPTS();
266 0 : buf = read_stream_next_buffer(stream, NULL);
267 0 : ReleaseBuffer(buf);
268 0 : ++blocks_done;
269 0 : }
270 0 : Assert(read_stream_next_buffer(stream, NULL) == InvalidBuffer);
271 0 : read_stream_end(stream);
272 0 : }
273 :
274 : /* Close relation, release locks. */
275 0 : relation_close(rel, AccessShareLock);
276 :
277 0 : if (privOid != relOid)
278 0 : UnlockRelationOid(privOid, AccessShareLock);
279 :
280 0 : PG_RETURN_INT64(blocks_done);
281 0 : }
|
