|
| 1 | +/*------------------------------------------------------------------------- |
| 2 | + * |
| 3 | + * tsm_system_rows.c |
| 4 | + * interface routines for system_rows tablesample method |
| 5 | + * |
| 6 | + * |
| 7 | + * Portions Copyright (c) 1996-2014, PostgreSQL Global Development Group |
| 8 | + * |
| 9 | + * IDENTIFICATION |
| 10 | + * contrib/tsm_system_rows_rowlimit/tsm_system_rows.c |
| 11 | + * |
| 12 | + *------------------------------------------------------------------------- |
| 13 | + */ |
| 14 | + |
| 15 | +#include"postgres.h" |
| 16 | + |
| 17 | +#include"fmgr.h" |
| 18 | + |
| 19 | +#include"access/tablesample.h" |
| 20 | +#include"access/relscan.h" |
| 21 | +#include"miscadmin.h" |
| 22 | +#include"nodes/execnodes.h" |
| 23 | +#include"nodes/relation.h" |
| 24 | +#include"optimizer/clauses.h" |
| 25 | +#include"storage/bufmgr.h" |
| 26 | +#include"utils/sampling.h" |
| 27 | + |
| 28 | +PG_MODULE_MAGIC; |
| 29 | + |
| 30 | +/* |
| 31 | + * State |
| 32 | + */ |
| 33 | +typedefstruct |
| 34 | +{ |
| 35 | +SamplerRandomStaterandstate; |
| 36 | +uint32seed;/* random seed */ |
| 37 | +BlockNumbernblocks;/* number of block in relation */ |
| 38 | +int32ntuples;/* number of tuples to return */ |
| 39 | +int32donetuples;/* tuples already returned */ |
| 40 | +OffsetNumberlt;/* last tuple returned from current block */ |
| 41 | +BlockNumberstep;/* step size */ |
| 42 | +BlockNumberlb;/* last block visited */ |
| 43 | +BlockNumberdoneblocks;/* number of already returned blocks */ |
| 44 | +}SystemSamplerData; |
| 45 | + |
| 46 | + |
| 47 | +PG_FUNCTION_INFO_V1(tsm_system_rows_init); |
| 48 | +PG_FUNCTION_INFO_V1(tsm_system_rows_nextblock); |
| 49 | +PG_FUNCTION_INFO_V1(tsm_system_rows_nexttuple); |
| 50 | +PG_FUNCTION_INFO_V1(tsm_system_rows_examinetuple); |
| 51 | +PG_FUNCTION_INFO_V1(tsm_system_rows_end); |
| 52 | +PG_FUNCTION_INFO_V1(tsm_system_rows_reset); |
| 53 | +PG_FUNCTION_INFO_V1(tsm_system_rows_cost); |
| 54 | + |
| 55 | +staticuint32random_relative_prime(uint32n,SamplerRandomStaterandstate); |
| 56 | + |
| 57 | +/* |
| 58 | + * Initializes the state. |
| 59 | + */ |
| 60 | +Datum |
| 61 | +tsm_system_rows_init(PG_FUNCTION_ARGS) |
| 62 | +{ |
| 63 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 64 | +uint32seed=PG_GETARG_UINT32(1); |
| 65 | +int32ntuples=PG_ARGISNULL(2) ?-1 :PG_GETARG_INT32(2); |
| 66 | +HeapScanDescscan=tsdesc->heapScan; |
| 67 | +SystemSamplerData*sampler; |
| 68 | + |
| 69 | +if (ntuples<1) |
| 70 | +ereport(ERROR, |
| 71 | +(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE), |
| 72 | +errmsg("invalid sample size"), |
| 73 | +errhint("Sample size must be positive integer value."))); |
| 74 | + |
| 75 | +sampler=palloc0(sizeof(SystemSamplerData)); |
| 76 | + |
| 77 | +/* Remember initial values for reinit */ |
| 78 | +sampler->seed=seed; |
| 79 | +sampler->nblocks=scan->rs_nblocks; |
| 80 | +sampler->ntuples=ntuples; |
| 81 | +sampler->donetuples=0; |
| 82 | +sampler->lt=InvalidOffsetNumber; |
| 83 | +sampler->doneblocks=0; |
| 84 | + |
| 85 | +sampler_random_init_state(sampler->seed,sampler->randstate); |
| 86 | + |
| 87 | +/* Find relative prime as step size for linear probing. */ |
| 88 | +sampler->step=random_relative_prime(sampler->nblocks,sampler->randstate); |
| 89 | +/* |
| 90 | + * Randomize start position so that blocks close to step size don't have |
| 91 | + * higher probability of being chosen on very short scan. |
| 92 | + */ |
| 93 | +sampler->lb=sampler_random_fract(sampler->randstate)* |
| 94 | +(sampler->nblocks /sampler->step); |
| 95 | + |
| 96 | +tsdesc->tsmdata= (void*)sampler; |
| 97 | + |
| 98 | +PG_RETURN_VOID(); |
| 99 | +} |
| 100 | + |
| 101 | +/* |
| 102 | + * Get next block number or InvalidBlockNumber when we're done. |
| 103 | + * |
| 104 | + * Uses linear probing algorithm for picking next block. |
| 105 | + */ |
| 106 | +Datum |
| 107 | +tsm_system_rows_nextblock(PG_FUNCTION_ARGS) |
| 108 | +{ |
| 109 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 110 | +SystemSamplerData*sampler= (SystemSamplerData*)tsdesc->tsmdata; |
| 111 | + |
| 112 | +sampler->lb= (sampler->lb+sampler->step) %sampler->nblocks; |
| 113 | +sampler->doneblocks++; |
| 114 | + |
| 115 | +/* All blocks have been read, we're done */ |
| 116 | +if (sampler->doneblocks>sampler->nblocks|| |
| 117 | +sampler->donetuples >=sampler->ntuples) |
| 118 | +PG_RETURN_UINT32(InvalidBlockNumber); |
| 119 | + |
| 120 | +PG_RETURN_UINT32(sampler->lb); |
| 121 | +} |
| 122 | + |
| 123 | +/* |
| 124 | + * Get next tuple offset in current block or InvalidOffsetNumber if we are done |
| 125 | + * with this block. |
| 126 | + */ |
| 127 | +Datum |
| 128 | +tsm_system_rows_nexttuple(PG_FUNCTION_ARGS) |
| 129 | +{ |
| 130 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 131 | +OffsetNumbermaxoffset=PG_GETARG_UINT16(2); |
| 132 | +SystemSamplerData*sampler= (SystemSamplerData*)tsdesc->tsmdata; |
| 133 | +OffsetNumbertupoffset=sampler->lt; |
| 134 | + |
| 135 | +if (tupoffset==InvalidOffsetNumber) |
| 136 | +tupoffset=FirstOffsetNumber; |
| 137 | +else |
| 138 | +tupoffset++; |
| 139 | + |
| 140 | +if (tupoffset>maxoffset|| |
| 141 | +sampler->donetuples >=sampler->ntuples) |
| 142 | +tupoffset=InvalidOffsetNumber; |
| 143 | + |
| 144 | +sampler->lt=tupoffset; |
| 145 | + |
| 146 | +PG_RETURN_UINT16(tupoffset); |
| 147 | +} |
| 148 | + |
| 149 | +/* |
| 150 | + * Examine tuple and decide if it should be returned. |
| 151 | + */ |
| 152 | +Datum |
| 153 | +tsm_system_rows_examinetuple(PG_FUNCTION_ARGS) |
| 154 | +{ |
| 155 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 156 | +boolvisible=PG_GETARG_BOOL(3); |
| 157 | +SystemSamplerData*sampler= (SystemSamplerData*)tsdesc->tsmdata; |
| 158 | + |
| 159 | +if (!visible) |
| 160 | +PG_RETURN_BOOL(false); |
| 161 | + |
| 162 | +sampler->donetuples++; |
| 163 | + |
| 164 | +PG_RETURN_BOOL(true); |
| 165 | +} |
| 166 | + |
| 167 | +/* |
| 168 | + * Cleanup method. |
| 169 | + */ |
| 170 | +Datum |
| 171 | +tsm_system_rows_end(PG_FUNCTION_ARGS) |
| 172 | +{ |
| 173 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 174 | + |
| 175 | +pfree(tsdesc->tsmdata); |
| 176 | + |
| 177 | +PG_RETURN_VOID(); |
| 178 | +} |
| 179 | + |
| 180 | +/* |
| 181 | + * Reset state (called by ReScan). |
| 182 | + */ |
| 183 | +Datum |
| 184 | +tsm_system_rows_reset(PG_FUNCTION_ARGS) |
| 185 | +{ |
| 186 | +TableSampleDesc*tsdesc= (TableSampleDesc*)PG_GETARG_POINTER(0); |
| 187 | +SystemSamplerData*sampler= (SystemSamplerData*)tsdesc->tsmdata; |
| 188 | + |
| 189 | +sampler->lt=InvalidOffsetNumber; |
| 190 | +sampler->donetuples=0; |
| 191 | +sampler->doneblocks=0; |
| 192 | + |
| 193 | +sampler_random_init_state(sampler->seed,sampler->randstate); |
| 194 | +sampler->step=random_relative_prime(sampler->nblocks,sampler->randstate); |
| 195 | +sampler->lb=sampler_random_fract(sampler->randstate)* (sampler->nblocks /sampler->step); |
| 196 | + |
| 197 | +PG_RETURN_VOID(); |
| 198 | +} |
| 199 | + |
| 200 | +/* |
| 201 | + * Costing function. |
| 202 | + */ |
| 203 | +Datum |
| 204 | +tsm_system_rows_cost(PG_FUNCTION_ARGS) |
| 205 | +{ |
| 206 | +PlannerInfo*root= (PlannerInfo*)PG_GETARG_POINTER(0); |
| 207 | +Path*path= (Path*)PG_GETARG_POINTER(1); |
| 208 | +RelOptInfo*baserel= (RelOptInfo*)PG_GETARG_POINTER(2); |
| 209 | +List*args= (List*)PG_GETARG_POINTER(3); |
| 210 | +BlockNumber*pages= (BlockNumber*)PG_GETARG_POINTER(4); |
| 211 | +double*tuples= (double*)PG_GETARG_POINTER(5); |
| 212 | +Node*limitnode; |
| 213 | +int32ntuples; |
| 214 | + |
| 215 | +limitnode=linitial(args); |
| 216 | +limitnode=estimate_expression_value(root,limitnode); |
| 217 | + |
| 218 | +if (IsA(limitnode,RelabelType)) |
| 219 | +limitnode= (Node*) ((RelabelType*)limitnode)->arg; |
| 220 | + |
| 221 | +if (IsA(limitnode,Const)) |
| 222 | +ntuples=DatumGetInt32(((Const*)limitnode)->constvalue); |
| 223 | +else |
| 224 | +{ |
| 225 | +/* Default ntuples if the estimation didn't return Const. */ |
| 226 | +ntuples=1000; |
| 227 | +} |
| 228 | + |
| 229 | +*pages=Min(baserel->pages,ntuples); |
| 230 | +*tuples=ntuples; |
| 231 | +path->rows=*tuples; |
| 232 | + |
| 233 | +PG_RETURN_VOID(); |
| 234 | +} |
| 235 | + |
| 236 | + |
| 237 | +staticuint32 |
| 238 | +gcd (uint32a,uint32b) |
| 239 | +{ |
| 240 | +uint32c; |
| 241 | + |
| 242 | +while (a!=0) |
| 243 | +{ |
| 244 | +c=a; |
| 245 | +a=b %a; |
| 246 | +b=c; |
| 247 | +} |
| 248 | + |
| 249 | +returnb; |
| 250 | +} |
| 251 | + |
| 252 | +staticuint32 |
| 253 | +random_relative_prime(uint32n,SamplerRandomStaterandstate) |
| 254 | +{ |
| 255 | +/* Pick random starting number, with some limits on what it can be. */ |
| 256 | +uint32r= (uint32)sampler_random_fract(randstate)*n/2+n/4, |
| 257 | +t; |
| 258 | + |
| 259 | +/* |
| 260 | + * This should only take 2 or 3 iterations as the probability of 2 numbers |
| 261 | + * being relatively prime is ~61%. |
| 262 | + */ |
| 263 | +while ((t=gcd(r,n))>1) |
| 264 | +{ |
| 265 | +CHECK_FOR_INTERRUPTS(); |
| 266 | +r /=t; |
| 267 | +} |
| 268 | + |
| 269 | +returnr; |
| 270 | +} |