PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
logicalfuncs.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * logicalfuncs.c
4  *
5  * Support functions for using logical decoding and management of
6  * logical replication slots via SQL.
7  *
8  *
9  * Copyright (c) 2012-2017, PostgreSQL Global Development Group
10  *
11  * IDENTIFICATION
12  * src/backend/replication/logicalfuncs.c
13  *-------------------------------------------------------------------------
14  */
15 
16 #include "postgres.h"
17 
18 #include <unistd.h>
19 
20 #include "fmgr.h"
21 #include "funcapi.h"
22 #include "miscadmin.h"
23 
24 #include "access/xlog_internal.h"
25 #include "access/xlogutils.h"
26 
27 #include "access/xact.h"
28 
29 #include "catalog/pg_type.h"
30 
31 #include "nodes/makefuncs.h"
32 
33 #include "mb/pg_wchar.h"
34 
35 #include "utils/array.h"
36 #include "utils/builtins.h"
37 #include "utils/inval.h"
38 #include "utils/memutils.h"
39 #include "utils/pg_lsn.h"
40 #include "utils/regproc.h"
41 #include "utils/resowner.h"
42 #include "utils/lsyscache.h"
43 
44 #include "replication/decode.h"
45 #include "replication/logical.h"
47 #include "replication/message.h"
48 
49 #include "storage/fd.h"
50 
51 /* private date for writing out data */
52 typedef struct DecodingOutputState
53 {
59 
60 /*
61  * Prepare for an output plugin write.
62  */
63 static void
65  bool last_write)
66 {
67  resetStringInfo(ctx->out);
68 }
69 
70 /*
71  * Perform output plugin write into tuplestore.
72  */
73 static void
75  bool last_write)
76 {
77  Datum values[3];
78  bool nulls[3];
80 
81  /* SQL Datums can only be of a limited length... */
82  if (ctx->out->len > MaxAllocSize - VARHDRSZ)
83  elog(ERROR, "too much output for sql interface");
84 
86 
87  memset(nulls, 0, sizeof(nulls));
88  values[0] = LSNGetDatum(lsn);
89  values[1] = TransactionIdGetDatum(xid);
90 
91  /*
92  * Assert ctx->out is in database encoding when we're writing textual
93  * output.
94  */
95  if (!p->binary_output)
97  ctx->out->data, ctx->out->len,
98  false));
99 
100  /* ick, but cstring_to_text_with_len works for bytea perfectly fine */
101  values[2] = PointerGetDatum(
102  cstring_to_text_with_len(ctx->out->data, ctx->out->len));
103 
104  tuplestore_putvalues(p->tupstore, p->tupdesc, values, nulls);
105  p->returned_rows++;
106 }
107 
108 static void
110 {
111  if (!superuser() && !has_rolreplication(GetUserId()))
112  ereport(ERROR,
113  (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
114  (errmsg("must be superuser or replication role to use replication slots"))));
115 }
116 
117 int
119  int reqLen, XLogRecPtr targetRecPtr, char *cur_page, TimeLineID *pageTLI)
120 {
121  return read_local_xlog_page(state, targetPagePtr, reqLen,
122  targetRecPtr, cur_page, pageTLI);
123 }
124 
125 /*
126  * Helper function for the various SQL callable logical decoding functions.
127  */
128 static Datum
129 pg_logical_slot_get_changes_guts(FunctionCallInfo fcinfo, bool confirm, bool binary)
130 {
131  Name name;
132  XLogRecPtr upto_lsn;
133  int32 upto_nchanges;
134  ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
135  MemoryContext per_query_ctx;
136  MemoryContext oldcontext;
137  XLogRecPtr end_of_wal;
138  XLogRecPtr startptr;
140  ResourceOwner old_resowner = CurrentResourceOwner;
141  ArrayType *arr;
142  Size ndim;
143  List *options = NIL;
145 
147 
149 
150  if (PG_ARGISNULL(0))
151  ereport(ERROR,
152  (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
153  errmsg("slot name must not be null")));
154  name = PG_GETARG_NAME(0);
155 
156  if (PG_ARGISNULL(1))
157  upto_lsn = InvalidXLogRecPtr;
158  else
159  upto_lsn = PG_GETARG_LSN(1);
160 
161  if (PG_ARGISNULL(2))
162  upto_nchanges = InvalidXLogRecPtr;
163  else
164  upto_nchanges = PG_GETARG_INT32(2);
165 
166  if (PG_ARGISNULL(3))
167  ereport(ERROR,
168  (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
169  errmsg("options array must not be null")));
170  arr = PG_GETARG_ARRAYTYPE_P(3);
171 
172  /* check to see if caller supports us returning a tuplestore */
173  if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
174  ereport(ERROR,
175  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
176  errmsg("set-valued function called in context that cannot accept a set")));
177  if (!(rsinfo->allowedModes & SFRM_Materialize))
178  ereport(ERROR,
179  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
180  errmsg("materialize mode required, but it is not allowed in this context")));
181 
182  /* state to write output to */
183  p = palloc0(sizeof(DecodingOutputState));
184 
185  p->binary_output = binary;
186 
187  /* Build a tuple descriptor for our result type */
188  if (get_call_result_type(fcinfo, NULL, &p->tupdesc) != TYPEFUNC_COMPOSITE)
189  elog(ERROR, "return type must be a row type");
190 
191  per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
192  oldcontext = MemoryContextSwitchTo(per_query_ctx);
193 
194  /* Deconstruct options array */
195  ndim = ARR_NDIM(arr);
196  if (ndim > 1)
197  {
198  ereport(ERROR,
199  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
200  errmsg("array must be one-dimensional")));
201  }
202  else if (array_contains_nulls(arr))
203  {
204  ereport(ERROR,
205  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
206  errmsg("array must not contain nulls")));
207  }
208  else if (ndim == 1)
209  {
210  int nelems;
211  Datum *datum_opts;
212  int i;
213 
214  Assert(ARR_ELEMTYPE(arr) == TEXTOID);
215 
216  deconstruct_array(arr, TEXTOID, -1, false, 'i',
217  &datum_opts, NULL, &nelems);
218 
219  if (nelems % 2 != 0)
220  ereport(ERROR,
221  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
222  errmsg("array must have even number of elements")));
223 
224  for (i = 0; i < nelems; i += 2)
225  {
226  char *name = TextDatumGetCString(datum_opts[i]);
227  char *opt = TextDatumGetCString(datum_opts[i + 1]);
228 
229  options = lappend(options, makeDefElem(name, (Node *) makeString(opt), -1));
230  }
231  }
232 
233  p->tupstore = tuplestore_begin_heap(true, false, work_mem);
234  rsinfo->returnMode = SFRM_Materialize;
235  rsinfo->setResult = p->tupstore;
236  rsinfo->setDesc = p->tupdesc;
237 
238  /* compute the current end-of-wal */
239  if (!RecoveryInProgress())
240  end_of_wal = GetFlushRecPtr();
241  else
242  end_of_wal = GetXLogReplayRecPtr(NULL);
243 
245 
246  PG_TRY();
247  {
248  /* restart at slot's confirmed_flush */
250  options,
254 
255  MemoryContextSwitchTo(oldcontext);
256 
257  /*
258  * Check whether the output plugin writes textual output if that's
259  * what we need.
260  */
261  if (!binary &&
262  ctx->options.output_type !=OUTPUT_PLUGIN_TEXTUAL_OUTPUT)
263  ereport(ERROR,
264  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
265  errmsg("logical decoding output plugin \"%s\" produces binary output, but function \"%s\" expects textual data",
267  format_procedure(fcinfo->flinfo->fn_oid))));
268 
269  ctx->output_writer_private = p;
270 
271  /*
272  * Decoding of WAL must start at restart_lsn so that the entirety of
273  * xacts that committed after the slot's confirmed_flush can be
274  * accumulated into reorder buffers.
275  */
276  startptr = MyReplicationSlot->data.restart_lsn;
277 
279 
280  /* invalidate non-timetravel entries */
282 
283  while ((startptr != InvalidXLogRecPtr && startptr < end_of_wal) ||
284  (ctx->reader->EndRecPtr != InvalidXLogRecPtr && ctx->reader->EndRecPtr < end_of_wal))
285  {
286  XLogRecord *record;
287  char *errm = NULL;
288 
289  record = XLogReadRecord(ctx->reader, startptr, &errm);
290  if (errm)
291  elog(ERROR, "%s", errm);
292 
293  /*
294  * Now that we've set up the xlog reader state, subsequent calls
295  * pass InvalidXLogRecPtr to say "continue from last record"
296  */
297  startptr = InvalidXLogRecPtr;
298 
299  /*
300  * The {begin_txn,change,commit_txn}_wrapper callbacks above will
301  * store the description into our tuplestore.
302  */
303  if (record != NULL)
304  LogicalDecodingProcessRecord(ctx, ctx->reader);
305 
306  /* check limits */
307  if (upto_lsn != InvalidXLogRecPtr &&
308  upto_lsn <= ctx->reader->EndRecPtr)
309  break;
310  if (upto_nchanges != 0 &&
311  upto_nchanges <= p->returned_rows)
312  break;
314  }
315 
316  tuplestore_donestoring(tupstore);
317 
318  CurrentResourceOwner = old_resowner;
319 
320  /*
321  * Next time, start where we left off. (Hunting things, the family
322  * business..)
323  */
324  if (ctx->reader->EndRecPtr != InvalidXLogRecPtr && confirm)
325  {
326  LogicalConfirmReceivedLocation(ctx->reader->EndRecPtr);
327  /*
328  * If only the confirmed_flush_lsn has changed the slot won't get
329  * marked as dirty by the above. Callers on the walsender interface
330  * are expected to keep track of their own progress and don't need
331  * it written out. But SQL-interface users cannot specify their own
332  * start positions and it's harder for them to keep track of their
333  * progress, so we should make more of an effort to save it for them.
334  *
335  * Dirty the slot so it's written out at the next checkpoint. We'll
336  * still lose its position on crash, as documented, but it's better
337  * than always losing the position even on clean restart.
338  */
340  }
341 
342  /* free context, call shutdown callback */
343  FreeDecodingContext(ctx);
344 
347  }
348  PG_CATCH();
349  {
350  /* clear all timetravel entries */
352 
353  PG_RE_THROW();
354  }
355  PG_END_TRY();
356 
357  return (Datum) 0;
358 }
359 
360 /*
361  * SQL function returning the changestream as text, consuming the data.
362  */
363 Datum
365 {
366  return pg_logical_slot_get_changes_guts(fcinfo, true, false);
367 }
368 
369 /*
370  * SQL function returning the changestream as text, only peeking ahead.
371  */
372 Datum
374 {
375  return pg_logical_slot_get_changes_guts(fcinfo, false, false);
376 }
377 
378 /*
379  * SQL function returning the changestream in binary, consuming the data.
380  */
381 Datum
383 {
384  return pg_logical_slot_get_changes_guts(fcinfo, true, true);
385 }
386 
387 /*
388  * SQL function returning the changestream in binary, only peeking ahead.
389  */
390 Datum
392 {
393  return pg_logical_slot_get_changes_guts(fcinfo, false, true);
394 }
395 
396 
397 /*
398  * SQL function for writing logical decoding message into WAL.
399  */
400 Datum
402 {
403  bool transactional = PG_GETARG_BOOL(0);
404  char *prefix = text_to_cstring(PG_GETARG_TEXT_PP(1));
405  bytea *data = PG_GETARG_BYTEA_PP(2);
406  XLogRecPtr lsn;
407 
408  lsn = LogLogicalMessage(prefix, VARDATA_ANY(data), VARSIZE_ANY_EXHDR(data),
409  transactional);
410  PG_RETURN_LSN(lsn);
411 }
412 
413 Datum
415 {
416  /* bytea and text are compatible */
417  return pg_logical_emit_message_bytea(fcinfo);
418 }
void tuplestore_putvalues(Tuplestorestate *state, TupleDesc tdesc, Datum *values, bool *isnull)
Definition: tuplestore.c:735
Value * makeString(char *str)
Definition: value.c:53
#define NIL
Definition: pg_list.h:69
#define PG_GETARG_INT32(n)
Definition: fmgr.h:225
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
#define IsA(nodeptr, _type_)
Definition: nodes.h:559
TypeFuncClass get_call_result_type(FunctionCallInfo fcinfo, Oid *resultTypeId, TupleDesc *resultTupleDesc)
Definition: funcapi.c:211
#define VARDATA_ANY(PTR)
Definition: postgres.h:349
uint32 TimeLineID
Definition: xlogdefs.h:45
uint32 TransactionId
Definition: c.h:394
int logical_read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, XLogRecPtr targetRecPtr, char *cur_page, TimeLineID *pageTLI)
Definition: logicalfuncs.c:118
Oid GetUserId(void)
Definition: miscinit.c:283
#define TEXTOID
Definition: pg_type.h:324
#define PointerGetDatum(X)
Definition: postgres.h:564
#define VARHDRSZ
Definition: c.h:441
ResourceOwner CurrentResourceOwner
Definition: resowner.c:138
static void check_permissions(void)
Definition: logicalfuncs.c:109
#define tuplestore_donestoring(state)
Definition: tuplestore.h:60
Datum pg_logical_emit_message_bytea(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:401
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
Definition: nodes.h:508
int errcode(int sqlerrcode)
Definition: elog.c:575
#define LSNGetDatum(X)
Definition: pg_lsn.h:22
bool superuser(void)
Definition: superuser.c:47
Datum pg_logical_slot_get_binary_changes(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:382
Tuplestorestate * tupstore
Definition: logicalfuncs.c:54
XLogRecPtr GetFlushRecPtr(void)
Definition: xlog.c:8154
#define PG_GETARG_BOOL(n)
Definition: fmgr.h:230
ReplicationSlotPersistentData data
Definition: slot.h:115
bool RecoveryInProgress(void)
Definition: xlog.c:7805
DefElem * makeDefElem(char *name, Node *arg, int location)
Definition: makefuncs.c:543
void InvalidateSystemCaches(void)
Definition: inval.c:634
XLogRecord * XLogReadRecord(XLogReaderState *state, XLogRecPtr RecPtr, char **errormsg)
Definition: xlogreader.c:193
signed int int32
Definition: c.h:253
#define PG_GETARG_TEXT_PP(n)
Definition: fmgr.h:270
#define PG_RETURN_LSN(x)
Definition: pg_lsn.h:25
int read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, XLogRecPtr targetRecPtr, char *cur_page, TimeLineID *pageTLI)
Definition: xlogutils.c:765
FmgrInfo * flinfo
Definition: fmgr.h:71
#define PG_GETARG_ARRAYTYPE_P(n)
Definition: array.h:244
static Datum pg_logical_slot_get_changes_guts(FunctionCallInfo fcinfo, bool confirm, bool binary)
Definition: logicalfuncs.c:129
#define ERROR
Definition: elog.h:43
void LogicalDecodingProcessRecord(LogicalDecodingContext *ctx, XLogReaderState *record)
Definition: decode.c:93
struct DecodingOutputState DecodingOutputState
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
Definition: xlog.c:10984
Definition: c.h:489
LogicalDecodingContext * CreateDecodingContext(XLogRecPtr start_lsn, List *output_plugin_options, XLogPageReadCB read_page, LogicalOutputPluginWriterPrepareWrite prepare_write, LogicalOutputPluginWriterWrite do_write)
Definition: logical.c:327
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:163
fmNodePtr resultinfo
Definition: fmgr.h:73
void resetStringInfo(StringInfo str)
Definition: stringinfo.c:94
void ReplicationSlotRelease(void)
Definition: slot.c:374
static void LogicalOutputPrepareWrite(LogicalDecodingContext *ctx, XLogRecPtr lsn, TransactionId xid, bool last_write)
Definition: logicalfuncs.c:64
bool pg_verify_mbstr(int encoding, const char *mbstr, int len, bool noError)
Definition: wchar.c:1877
#define ereport(elevel, rest)
Definition: elog.h:122
List * lappend(List *list, void *datum)
Definition: list.c:128
XLogRecPtr LogLogicalMessage(const char *prefix, const char *message, size_t size, bool transactional)
Definition: message.c:51
#define MaxAllocSize
Definition: memutils.h:40
Tuplestorestate * tuplestore_begin_heap(bool randomAccess, bool interXact, int maxKBytes)
Definition: tuplestore.c:316
#define TextDatumGetCString(d)
Definition: builtins.h:91
#define TransactionIdGetDatum(X)
Definition: postgres.h:529
void * palloc0(Size size)
Definition: mcxt.c:920
#define PG_GETARG_LSN(n)
Definition: pg_lsn.h:24
uintptr_t Datum
Definition: postgres.h:374
int GetDatabaseEncoding(void)
Definition: mbutils.c:1015
int work_mem
Definition: globals.c:112
Oid fn_oid
Definition: fmgr.h:56
int allowedModes
Definition: execnodes.h:199
char * format_procedure(Oid procedure_oid)
Definition: regproc.c:367
Datum pg_logical_slot_get_changes(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:364
SetFunctionReturnMode returnMode
Definition: execnodes.h:201
#define PG_CATCH()
Definition: elog.h:293
ReplicationSlot * MyReplicationSlot
Definition: slot.c:95
#define PG_ARGISNULL(n)
Definition: fmgr.h:166
#define NULL
Definition: c.h:226
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:671
Definition: regguts.h:298
void FreeDecodingContext(LogicalDecodingContext *ctx)
Definition: logical.c:457
XLogRecPtr restart_lsn
Definition: slot.h:68
void ReplicationSlotAcquire(const char *name)
Definition: slot.c:325
size_t Size
Definition: c.h:353
#define PG_GETARG_BYTEA_PP(n)
Definition: fmgr.h:268
bool has_rolreplication(Oid roleid)
Definition: miscinit.c:464
#define PG_RE_THROW()
Definition: elog.h:314
void LogicalConfirmReceivedLocation(XLogRecPtr lsn)
Definition: logical.c:884
MemoryContext ecxt_per_query_memory
Definition: execnodes.h:133
#define ARR_NDIM(a)
Definition: array.h:271
const char * name
Definition: encode.c:521
Tuplestorestate * setResult
Definition: execnodes.h:204
void deconstruct_array(ArrayType *array, Oid elmtype, int elmlen, bool elmbyval, char elmalign, Datum **elemsp, bool **nullsp, int *nelemsp)
Definition: arrayfuncs.c:3475
static Datum values[MAXATTR]
Definition: bootstrap.c:162
char * text_to_cstring(const text *t)
Definition: varlena.c:184
ExprContext * econtext
Definition: execnodes.h:197
TupleDesc setDesc
Definition: execnodes.h:205
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:342
int errmsg(const char *fmt,...)
Definition: elog.c:797
StringInfo out
Definition: logical.h:57
int i
#define NameStr(name)
Definition: c.h:495
Datum pg_logical_slot_peek_changes(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:373
Definition: c.h:435
#define PG_FUNCTION_ARGS
Definition: fmgr.h:150
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:97
#define elog
Definition: elog.h:219
#define PG_TRY()
Definition: elog.h:284
bool array_contains_nulls(ArrayType *array)
Definition: arrayfuncs.c:3542
Definition: pg_list.h:45
void CheckLogicalDecodingRequirements(void)
Definition: logical.c:76
#define ARR_ELEMTYPE(a)
Definition: array.h:273
void * output_writer_private
Definition: logical.h:67
#define PG_END_TRY()
Definition: elog.h:300
static void LogicalOutputWrite(LogicalDecodingContext *ctx, XLogRecPtr lsn, TransactionId xid, bool last_write)
Definition: logicalfuncs.c:74
void ReplicationSlotMarkDirty(void)
Definition: slot.c:580
#define PG_GETARG_NAME(n)
Definition: fmgr.h:234
Datum pg_logical_emit_message_text(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:414
ResourceOwner ResourceOwnerCreate(ResourceOwner parent, const char *name)
Definition: resowner.c:416
Datum pg_logical_slot_peek_binary_changes(PG_FUNCTION_ARGS)
Definition: logicalfuncs.c:391