#include "postgres.h"
#include "pgstat.h"
+#include "access/transam.h"
#include "catalog/namespace.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
#include "nodes/parsenodes.h"
#include "storage/bufmgr.h"
#include "utils/builtins.h"
+#include "utils/hsearch.h"
+#include "utils/memutils.h"
#include "utils/lsyscache.h"
#include "utils/guc.h"
#if PG_VERSION_NUM >= 90200
#include "utils/timestamp.h"
#if PG_VERSION_NUM >= 90500
#include "nodes/makefuncs.h"
+#if PG_VERSION_NUM >= 100000
+#include "utils/varlena.h"
+#include "utils/regproc.h"
+#endif
#endif
#endif
static bool online_analyze_verbose = true;
static double online_analyze_scale_factor = 0.1;
static int online_analyze_threshold = 50;
+static int online_analyze_capacity_threshold = 100000;
static double online_analyze_min_interval = 10000;
+static int online_analyze_lower_limit = 0;
static ExecutorEnd_hook_type oldExecutorEndHook = NULL;
#if PG_VERSION_NUM >= 90200
static ProcessUtility_hook_type oldProcessUtilityHook = NULL;
#endif
+typedef enum CmdKind
+{
+ CK_SELECT = CMD_SELECT,
+ CK_UPDATE = CMD_UPDATE,
+ CK_INSERT = CMD_INSERT,
+ CK_DELETE = CMD_DELETE,
+ CK_TRUNCATE,
+ CK_FASTTRUNCATE,
+ CK_CREATE
+} CmdKind;
+
+
typedef enum
{
OATT_ALL = 0x03,
static TableList excludeTables = {0, NULL, NULL};
static TableList includeTables = {0, NULL, NULL};
+typedef struct OnlineAnalyzeTableStat {
+ Oid tableid;
+ bool rereadStat;
+ PgStat_Counter n_tuples;
+ PgStat_Counter changes_since_analyze;
+ TimestampTz autovac_analyze_timestamp;
+ TimestampTz analyze_timestamp;
+} OnlineAnalyzeTableStat;
+
+static MemoryContext onlineAnalyzeMemoryContext = NULL;
+static HTAB *relstats = NULL;
+
+static void relstatsInit(void);
+
+#if PG_VERSION_NUM < 100000
static int
oid_cmp(const void *a, const void *b)
{
return 0;
return (*(Oid*)a > *(Oid*)b) ? 1 : -1;
}
+#endif
static const char *
tableListAssign(const char * newval, bool doit, TableList *tbl)
#endif
static void
-makeAnalyze(Oid relOid, CmdType operation, int32 naffected)
+makeAnalyze(Oid relOid, CmdKind operation, int64 naffected)
{
- PgStat_StatTabEntry *tabentry;
TimestampTz now = GetCurrentTimestamp();
+ Relation rel;
+ OnlineAnalyzeTableType reltype;
+ bool found = false,
+ newTable = false;
+ OnlineAnalyzeTableStat *rstat,
+ dummyrstat;
+ PgStat_StatTabEntry *tabentry = NULL;
if (relOid == InvalidOid)
return;
+ elog(NOTICE,"makeAnalyze operation: %d naffected: %d",
+ operation, (int32)naffected);
+
if (naffected == 0)
- /* return if there is not changes */
+ /* return if there is no changes */
return;
else if (naffected < 0)
/* number if affected rows is unknown */
naffected = 0;
- if (get_rel_relkind(relOid) != RELKIND_RELATION)
+ rel = RelationIdGetRelation(relOid);
+ if (rel->rd_rel->relkind != RELKIND_RELATION)
+ {
+ RelationClose(rel);
return;
+ }
+
+ reltype =
+#if PG_VERSION_NUM >= 90100
+ (rel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
+#else
+ (rel->rd_istemp || rel->rd_islocaltemp)
+#endif
+ ? OATT_TEMPORARY : OATT_PERSISTENT;
- tabentry = pgstat_fetch_stat_tabentry(relOid);
+ RelationClose(rel);
+ /*
+ * includeTables overwrites excludeTables
+ */
+ switch(online_analyze_table_type)
+ {
+ case OATT_ALL:
+ if (get_rel_relkind(relOid) != RELKIND_RELATION ||
+ (matchOid(&excludeTables, relOid) == true &&
+ matchOid(&includeTables, relOid) == false))
+ return;
+ break;
+ case OATT_NONE:
+ if (get_rel_relkind(relOid) != RELKIND_RELATION ||
+ matchOid(&includeTables, relOid) == false)
+ return;
+ break;
+ case OATT_TEMPORARY:
+ case OATT_PERSISTENT:
+ default:
+ /*
+ * skip analyze if relation's type doesn't not match
+ * online_analyze_table_type
+ */
+ if ((online_analyze_table_type & reltype) == 0 ||
+ matchOid(&excludeTables, relOid) == true)
+ {
+ if (matchOid(&includeTables, relOid) == false)
+ return;
+ }
+ break;
+ }
+
+ /*
+ * Do not store data about persistent table in local memory because we
+ * could not track changes of them: they could be changed by another
+ * backends. So always get a pgstat table entry.
+ */
+ if (reltype == OATT_TEMPORARY)
+ rstat = hash_search(relstats, &relOid, HASH_ENTER, &found);
+ else
+ rstat = &dummyrstat; /* found == false for following if */
+
+ if (!found)
+ {
+ MemSet(rstat, 0, sizeof(*rstat));
+ rstat->tableid = relOid;
+ newTable = true;
+ }
+
+ Assert(rstat->tableid == relOid);
+
+ /* do not rered data if it was a truncation */
+ if (operation != CK_TRUNCATE && operation != CK_FASTTRUNCATE &&
+ (newTable == true || rstat->rereadStat == true))
+ {
+ rstat->rereadStat = false;
+
+ tabentry = pgstat_fetch_stat_tabentry(relOid);
+
+ if (tabentry)
+ {
+ rstat->n_tuples = tabentry->n_dead_tuples + tabentry->n_live_tuples;
+ rstat->changes_since_analyze =
#if PG_VERSION_NUM >= 90000
-#define changes_since_analyze(t) ((t)->changes_since_analyze)
+ tabentry->changes_since_analyze;
#else
-#define changes_since_analyze(t) ((t)->n_live_tuples + (t)->n_dead_tuples - (t)->last_anl_tuples)
-#endif
-
- if (
- tabentry == NULL /* a new table */ ||
- (
- /* do not analyze too often, if both stamps are exceeded the go */
- TimestampDifferenceExceeds(tabentry->analyze_timestamp, now, online_analyze_min_interval) &&
- TimestampDifferenceExceeds(tabentry->autovac_analyze_timestamp, now, online_analyze_min_interval) &&
- /* be in sync with relation_needs_vacanalyze */
- ((double)(changes_since_analyze(tabentry) + naffected)) >=
- online_analyze_scale_factor * ((double)(tabentry->n_dead_tuples + tabentry->n_live_tuples)) +
- (double)online_analyze_threshold
- )
- )
+ tabentry->n_live_tuples + tabentry->n_dead_tuples -
+ tabentry->last_anl_tuples;
+#endif
+ rstat->autovac_analyze_timestamp =
+ tabentry->autovac_analyze_timestamp;
+ rstat->analyze_timestamp = tabentry->analyze_timestamp;
+ }
+ }
+
+ if (newTable ||
+ /* force analyze after truncate, fasttruncate already did analyze */
+ operation == CK_TRUNCATE || (
+ /* do not analyze too often, if both stamps are exceeded the go */
+ TimestampDifferenceExceeds(rstat->analyze_timestamp, now, online_analyze_min_interval) &&
+ TimestampDifferenceExceeds(rstat->autovac_analyze_timestamp, now, online_analyze_min_interval) &&
+ /* do not analyze too small tables */
+ rstat->n_tuples + rstat->changes_since_analyze + naffected > online_analyze_lower_limit &&
+ /* be in sync with relation_needs_vacanalyze */
+ ((double)(rstat->changes_since_analyze + naffected)) >=
+ online_analyze_scale_factor * ((double)rstat->n_tuples) +
+ (double)online_analyze_threshold))
{
#if PG_VERSION_NUM < 90500
VacuumStmt vacstmt;
#endif
TimestampTz startStamp, endStamp;
- memset(&startStamp, 0, sizeof(startStamp)); /* keep compiler quiet */
- /*
- * includeTables overwrites excludeTables
- */
- switch(online_analyze_table_type)
- {
- case OATT_ALL:
- if (matchOid(&excludeTables, relOid) == true &&
- matchOid(&includeTables, relOid) == false)
- return;
- break;
- case OATT_NONE:
- if (matchOid(&includeTables, relOid) == false)
- return;
- break;
- case OATT_TEMPORARY:
- case OATT_PERSISTENT:
- default:
- {
- Relation rel;
- OnlineAnalyzeTableType reltype;
-
- rel = RelationIdGetRelation(relOid);
- reltype =
-#if PG_VERSION_NUM >= 90100
- (rel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
-#else
- (rel->rd_istemp || rel->rd_islocaltemp)
-#endif
- ? OATT_TEMPORARY : OATT_PERSISTENT;
- RelationClose(rel);
-
- /*
- * skip analyze if relation's type doesn't not match
- * online_analyze_table_type
- */
- if ((online_analyze_table_type & reltype) == 0 ||
- matchOid(&excludeTables, relOid) == true)
- {
- if (matchOid(&includeTables, relOid) == false)
- return;
- }
- }
- break;
- }
+ memset(&startStamp, 0, sizeof(startStamp)); /* keep compiler quiet */
memset(&vacstmt, 0, sizeof(vacstmt));
((double)secs) + ((double)microsecs)/1.0e6);
}
+ rstat->autovac_analyze_timestamp = now;
+ rstat->changes_since_analyze = 0;
- if (tabentry == NULL)
+ switch(operation)
{
- /* new table */
- pgstat_clear_snapshot();
+ case CK_CREATE:
+ case CK_INSERT:
+ case CK_UPDATE:
+ rstat->n_tuples += naffected;
+ case CK_DELETE:
+ rstat->rereadStat = (reltype == OATT_PERSISTENT);
+ break;
+ case CK_TRUNCATE:
+ case CK_FASTTRUNCATE:
+ rstat->n_tuples = 0;
+ break;
+ default:
+ break;
}
- else
+
+ /* update last analyze timestamp in local memory of backend */
+ if (tabentry)
{
- /* update last analyze timestamp in local memory of backend */
tabentry->analyze_timestamp = now;
+ tabentry->changes_since_analyze = 0;
}
+#if 0
+ /* force reload stat for new table */
+ if (newTable)
+ pgstat_clear_snapshot();
+#endif
}
-#if PG_VERSION_NUM >= 90000
- else if (tabentry != NULL)
+ else
{
- tabentry->changes_since_analyze += naffected;
- }
+#if PG_VERSION_NUM >= 90000
+ if (tabentry)
+ tabentry->changes_since_analyze += naffected;
#endif
+ switch(operation)
+ {
+ case CK_CREATE:
+ case CK_INSERT:
+ rstat->changes_since_analyze += naffected;
+ rstat->n_tuples += naffected;
+ break;
+ case CK_UPDATE:
+ rstat->changes_since_analyze += 2 * naffected;
+ rstat->n_tuples += naffected;
+ case CK_DELETE:
+ rstat->changes_since_analyze += naffected;
+ break;
+ case CK_TRUNCATE:
+ case CK_FASTTRUNCATE:
+ rstat->changes_since_analyze = 0;
+ rstat->n_tuples = 0;
+ break;
+ default:
+ break;
+ }
+ }
+
+ /* Reset local cache if we are over limit */
+ if (hash_get_num_entries(relstats) > online_analyze_capacity_threshold)
+ relstatsInit();
}
+static Const*
+isFastTruncateCall(QueryDesc *queryDesc)
+{
+ TargetEntry *te;
+ FuncExpr *fe;
+ Const *constval;
+
+ if (!(
+ queryDesc->plannedstmt &&
+ queryDesc->operation == CMD_SELECT &&
+ queryDesc->plannedstmt->planTree &&
+ queryDesc->plannedstmt->planTree->targetlist &&
+ list_length(queryDesc->plannedstmt->planTree->targetlist) == 1 &&
+ IsA(linitial(queryDesc->plannedstmt->planTree->targetlist), TargetEntry)
+ ))
+ return NULL;
+
+ te = linitial(queryDesc->plannedstmt->planTree->targetlist);
+
+ if (!IsA(te, TargetEntry))
+ return NULL;
+
+ fe = (FuncExpr*)te->expr;
+
+ if (!(
+ fe && IsA(fe, FuncExpr) &&
+ fe->funcid >= FirstNormalObjectId &&
+ fe->funcretset == false &&
+ fe->funcresulttype == VOIDOID &&
+ fe->funcvariadic == false &&
+ list_length(fe->args) == 1 &&
+ IsA(linitial(fe->args), Const)
+ ))
+ return NULL;
+
+ constval = linitial(fe->args);
+
+ if (!(
+ IsA(constval,Const) &&
+ constval->consttype == TEXTOID &&
+ strcmp(get_func_name(fe->funcid), "fasttruncate") == 0
+ ))
+ return NULL;
+
+ return constval;
+}
+
+
+
extern PGDLLIMPORT void onlineAnalyzeHooker(QueryDesc *queryDesc);
void
onlineAnalyzeHooker(QueryDesc *queryDesc)
{
- uint32 naffected = -1;
+ int64 naffected = -1;
+ Const *constval;
if (queryDesc->estate)
naffected = queryDesc->estate->es_processed;
+#if PG_VERSION_NUM >= 90200
+ if (online_analyze_enable &&
+ (constval = isFastTruncateCall(queryDesc)) != NULL)
+ {
+ Datum tblnamed = constval->constvalue;
+ char *tblname = text_to_cstring(DatumGetTextP(tblnamed));
+ RangeVar *tblvar =
+ makeRangeVarFromNameList(stringToQualifiedNameList(tblname));
+
+ makeAnalyze(RangeVarGetRelid(tblvar,
+ NoLock,
+ false),
+ CK_FASTTRUNCATE, -1);
+ }
+#endif
+
if (online_analyze_enable && queryDesc->plannedstmt &&
(queryDesc->operation == CMD_INSERT ||
queryDesc->operation == CMD_UPDATE ||
RangeTblEntry *rte = list_nth(queryDesc->plannedstmt->rtable, n-1);
if (rte->rtekind == RTE_RELATION)
- makeAnalyze(rte->relid, queryDesc->operation, naffected);
+ makeAnalyze(rte->relid, (CmdKind)queryDesc->operation, naffected);
}
}
}
#if PG_VERSION_NUM >= 90200
static void
-onlineAnalyzeHookerUtility(Node *parsetree, const char *queryString,
+onlineAnalyzeHookerUtility(
+#if PG_VERSION_NUM >= 100000
+ PlannedStmt *pstmt,
+#else
+ Node *parsetree,
+#endif
+ const char *queryString,
#if PG_VERSION_NUM >= 90300
ProcessUtilityContext context, ParamListInfo params,
+#if PG_VERSION_NUM >= 100000
+ QueryEnvironment *queryEnv,
+#endif
#else
ParamListInfo params, bool isTopLevel,
#endif
DestReceiver *dest, char *completionTag) {
- RangeVar *tblname = NULL;
+ List *tblnames = NIL;
+ CmdKind op = CK_INSERT;
+#if PG_VERSION_NUM >= 100000
+ Node *parsetree = NULL;
+
+ if (pstmt->commandType == CMD_UTILITY)
+ parsetree = pstmt->utilityStmt;
+#endif
- if (IsA(parsetree, CreateTableAsStmt) && ((CreateTableAsStmt*)parsetree)->into)
- tblname = (RangeVar*)copyObject(((CreateTableAsStmt*)parsetree)->into->rel);
+ if (parsetree && online_analyze_enable)
+ {
+ if (IsA(parsetree, CreateTableAsStmt) &&
+ ((CreateTableAsStmt*)parsetree)->into)
+ {
+ tblnames =
+ list_make1((RangeVar*)copyObject(((CreateTableAsStmt*)parsetree)->into->rel));
+ op = CK_CREATE;
+ }
+ else if (IsA(parsetree, TruncateStmt))
+ {
+ tblnames = list_copy(((TruncateStmt*)parsetree)->relations);
+ op = CK_TRUNCATE;
+ }
+ }
+
+#if PG_VERSION_NUM >= 100000
+#define parsetree pstmt
+#endif
if (oldProcessUtilityHook)
oldProcessUtilityHook(parsetree, queryString,
#if PG_VERSION_NUM >= 90300
context, params,
+#if PG_VERSION_NUM >= 100000
+ queryEnv,
+#endif
#else
params, isTopLevel,
#endif
standard_ProcessUtility(parsetree, queryString,
#if PG_VERSION_NUM >= 90300
context, params,
+#if PG_VERSION_NUM >= 100000
+ queryEnv,
+#endif
#else
params, isTopLevel,
#endif
dest, completionTag);
- if (tblname) {
- Oid tblOid = RangeVarGetRelid(tblname, NoLock, true);
+#if PG_VERSION_NUM >= 100000
+#undef parsetree
+#endif
+
+ if (tblnames) {
+ ListCell *l;
- makeAnalyze(tblOid, CMD_INSERT, -1);
+ foreach(l, tblnames)
+ {
+ RangeVar *tblname = (RangeVar*)lfirst(l);
+ Oid tblOid = RangeVarGetRelid(tblname, NoLock, true);
+
+ makeAnalyze(tblOid, op, -1);
+ }
}
}
#endif
+static void
+relstatsInit(void)
+{
+ HASHCTL hash_ctl;
+ int flags = 0;
+
+ MemSet(&hash_ctl, 0, sizeof(hash_ctl));
+
+ hash_ctl.hash = oid_hash;
+ flags |= HASH_FUNCTION;
+
+ if (onlineAnalyzeMemoryContext)
+ {
+ Assert(relstats != NULL);
+ MemoryContextReset(onlineAnalyzeMemoryContext);
+ }
+ else
+ {
+ Assert(relstats == NULL);
+ onlineAnalyzeMemoryContext =
+ AllocSetContextCreate(CacheMemoryContext,
+ "online_analyze storage context",
+#if PG_VERSION_NUM < 90600
+ ALLOCSET_DEFAULT_MINSIZE,
+ ALLOCSET_DEFAULT_INITSIZE,
+ ALLOCSET_DEFAULT_MAXSIZE
+#else
+ ALLOCSET_DEFAULT_SIZES
+#endif
+ );
+ }
+
+ hash_ctl.hcxt = onlineAnalyzeMemoryContext;
+ flags |= HASH_CONTEXT;
+
+ hash_ctl.keysize = sizeof(Oid);
+
+ hash_ctl.entrysize = sizeof(OnlineAnalyzeTableStat);
+ flags |= HASH_ELEM;
+
+ relstats = hash_create("online_analyze storage", 1024, &hash_ctl, flags);
+}
+
void _PG_init(void);
void
_PG_init(void)
{
+ relstatsInit();
+
oldExecutorEndHook = ExecutorEnd_hook;
ExecutorEnd_hook = onlineAnalyzeHooker;
NULL
);
+ DefineCustomIntVariable(
+ "online_analyze.capacity_threshold",
+ "Max local cache table capacity",
+ "Max local cache table capacity",
+ &online_analyze_capacity_threshold,
+#if PG_VERSION_NUM >= 80400
+ online_analyze_capacity_threshold,
+#endif
+ 0,
+ 0x7fffffff,
+ PGC_USERSET,
+#if PG_VERSION_NUM >= 80400
+ GUC_NOT_IN_SAMPLE,
+#if PG_VERSION_NUM >= 90100
+ NULL,
+#endif
+#endif
+ NULL,
+ NULL
+ );
+
DefineCustomRealVariable(
"online_analyze.min_interval",
"minimum time interval between analyze call (in milliseconds)",
#endif
includeTablesShow
);
+
+ DefineCustomIntVariable(
+ "online_analyze.lower_limit",
+ "min number of rows in table to analyze",
+ "min number of rows in table to analyze",
+ &online_analyze_lower_limit,
+#if PG_VERSION_NUM >= 80400
+ online_analyze_lower_limit,
+#endif
+ 0,
+ 0x7fffffff,
+ PGC_USERSET,
+#if PG_VERSION_NUM >= 80400
+ GUC_NOT_IN_SAMPLE,
+#if PG_VERSION_NUM >= 90100
+ NULL,
+#endif
+#endif
+ NULL,
+ NULL
+ );
+
}
void _PG_fini(void);