#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
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 ExecutorEnd_hook_type oldExecutorEndHook = NULL;
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);
+
static int
oid_cmp(const void *a, const void *b)
{
static void
makeAnalyze(Oid relOid, CmdType operation, int32 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;
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;
+ 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;
+
+ RelationClose(rel);
+
/*
* includeTables overwrites excludeTables
*/
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)
{
- Relation rel;
- OnlineAnalyzeTableType reltype;
-
- rel = RelationIdGetRelation(relOid);
-
- if (rel->rd_rel->relkind != RELKIND_RELATION)
- {
- RelationClose(rel);
+ if (matchOid(&includeTables, relOid) == false)
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;
- 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;
}
+ break;
}
- tabentry = pgstat_fetch_stat_tabentry(relOid);
+ /*
+ * 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 == false || rstat->rereadStat == true || naffected == 0)
+ {
+
+ if (!found)
+ {
+ MemSet(rstat, 0, sizeof(*rstat));
+ rstat->tableid = relOid;
+ }
+ Assert(rstat->tableid == relOid);
+
+ 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;
+ rstat->rereadStat = false;
+ }
+ else
+ {
+ newTable = true;
+ rstat->rereadStat = true;
+ }
+ }
+
+ if (newTable || (
+ /* 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) &&
+ /* 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 */
memset(&vacstmt, 0, sizeof(vacstmt));
((double)secs) + ((double)microsecs)/1.0e6);
}
- if (tabentry == NULL)
- {
- /* new table */
- pgstat_clear_snapshot();
- }
- else
+ rstat->autovac_analyze_timestamp = now;
+ rstat->changes_since_analyze = 0;
+ rstat->rereadStat = true;
+
+ /* 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
+ rstat->changes_since_analyze += naffected;
+ }
+
+ /* Reset local cache if we are over limit */
+ if (hash_get_num_entries(relstats) > online_analyze_capacity_threshold)
+ relstatsInit();
}
extern PGDLLIMPORT void onlineAnalyzeHooker(QueryDesc *queryDesc);
}
#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)",
projects / online_analyze.git / blobdiff