From 44a4d51a1295264b778a533a0a9615762803603d Mon Sep 17 00:00:00 2001 From: Tom Lane Date: Thu, 9 Jan 2025 17:14:19 -0500 Subject: [PATCH v2 2/4] Improve DecodeTimezoneAbbrev's caching logic. The previous implementation could only cope with caching a result found in the zoneabbrevtbl. It is worth expending a little more space to be able to cache results obtained from the IANA timezone data, especially since that's likely to be the majority use-case going forward. To do this, we have to reset the cache after a change in the timezone GUC not only the timezone_abbrev GUC, but that's not hard. In my testing, this puts the speed of repeated timestamptz_in calls back on par with what it was before the previous patch. Per report from Aleksander Alekseev and additional investigation. Discussion: https://postgr.es/m/CAJ7c6TOATjJqvhnYsui0=CO5XFMF4dvTGH+skzB--jNhqSQu5g@mail.gmail.com --- src/backend/commands/variable.c | 2 + src/backend/utils/adt/datetime.c | 69 ++++++++++++++++++++++++-------- src/include/utils/datetime.h | 2 + src/tools/pgindent/typedefs.list | 1 + 4 files changed, 58 insertions(+), 16 deletions(-) diff --git a/src/backend/commands/variable.c b/src/backend/commands/variable.c index 44796bf15a..4ad6e236d6 100644 --- a/src/backend/commands/variable.c +++ b/src/backend/commands/variable.c @@ -381,6 +381,8 @@ void assign_timezone(const char *newval, void *extra) { session_timezone = *((pg_tz **) extra); + /* datetime.c's cache of timezone abbrevs may now be obsolete */ + ClearTimeZoneAbbrevCache(); } /* diff --git a/src/backend/utils/adt/datetime.c b/src/backend/utils/adt/datetime.c index cb028d3934..ef04112602 100644 --- a/src/backend/utils/adt/datetime.c +++ b/src/backend/utils/adt/datetime.c @@ -259,7 +259,17 @@ static const datetkn *datecache[MAXDATEFIELDS] = {NULL}; static const datetkn *deltacache[MAXDATEFIELDS] = {NULL}; -static const datetkn *abbrevcache[MAXDATEFIELDS] = {NULL}; +/* Cache for results of timezone abbreviation lookups */ + +typedef struct TzAbbrevCache +{ + char abbrev[TOKMAXLEN + 1]; /* always NUL-terminated */ + char ftype; /* TZ, DTZ, or DYNTZ */ + int offset; /* GMT offset, if fixed-offset */ + pg_tz *tz; /* relevant zone, if variable-offset */ +} TzAbbrevCache; + +static TzAbbrevCache tzabbrevcache[MAXDATEFIELDS]; /* @@ -3126,15 +3136,28 @@ DecodeTimezoneAbbrev(int field, const char *lowtoken, int *ftype, int *offset, pg_tz **tz, DateTimeErrorExtra *extra) { + TzAbbrevCache *tzc = &tzabbrevcache[field]; bool isfixed; int isdst; const datetkn *tp; + /* + * Do we have a cached result? Use strncmp so that we match truncated + * names, although we shouldn't really see that happen with normal + * abbreviations. + */ + if (strncmp(lowtoken, tzc->abbrev, TOKMAXLEN) == 0) + { + *ftype = tzc->ftype; + *offset = tzc->offset; + *tz = tzc->tz; + return 0; + } + /* * See if the current session_timezone recognizes it. Checking this * before zoneabbrevtbl allows us to correctly handle abbreviations whose - * meaning varies across zones, such as "LMT". (Caching this lookup is - * left for later.) + * meaning varies across zones, such as "LMT". */ if (session_timezone && TimeZoneAbbrevIsKnown(lowtoken, session_timezone, @@ -3144,20 +3167,20 @@ DecodeTimezoneAbbrev(int field, const char *lowtoken, *tz = (isfixed ? NULL : session_timezone); /* flip sign to agree with the convention used in zoneabbrevtbl */ *offset = -(*offset); + /* cache result; use strlcpy to truncate name if necessary */ + strlcpy(tzc->abbrev, lowtoken, TOKMAXLEN + 1); + tzc->ftype = *ftype; + tzc->offset = *offset; + tzc->tz = *tz; return 0; } /* Nope, so look in zoneabbrevtbl */ - tp = abbrevcache[field]; - /* use strncmp so that we match truncated tokens */ - if (tp == NULL || strncmp(lowtoken, tp->token, TOKMAXLEN) != 0) - { - if (zoneabbrevtbl) - tp = datebsearch(lowtoken, zoneabbrevtbl->abbrevs, - zoneabbrevtbl->numabbrevs); - else - tp = NULL; - } + if (zoneabbrevtbl) + tp = datebsearch(lowtoken, zoneabbrevtbl->abbrevs, + zoneabbrevtbl->numabbrevs); + else + tp = NULL; if (tp == NULL) { *ftype = UNKNOWN_FIELD; @@ -3167,7 +3190,6 @@ DecodeTimezoneAbbrev(int field, const char *lowtoken, } else { - abbrevcache[field] = tp; *ftype = tp->type; if (tp->type == DYNTZ) { @@ -3181,11 +3203,26 @@ DecodeTimezoneAbbrev(int field, const char *lowtoken, *offset = tp->value; *tz = NULL; } + + /* cache result; use strlcpy to truncate name if necessary */ + strlcpy(tzc->abbrev, lowtoken, TOKMAXLEN + 1); + tzc->ftype = *ftype; + tzc->offset = *offset; + tzc->tz = *tz; } return 0; } +/* + * Reset tzabbrevcache after a change in session_timezone. + */ +void +ClearTimeZoneAbbrevCache(void) +{ + memset(tzabbrevcache, 0, sizeof(tzabbrevcache)); +} + /* DecodeSpecial() * Decode text string using lookup table. @@ -5036,8 +5073,8 @@ void InstallTimeZoneAbbrevs(TimeZoneAbbrevTable *tbl) { zoneabbrevtbl = tbl; - /* reset abbrevcache, which may contain pointers into old table */ - memset(abbrevcache, 0, sizeof(abbrevcache)); + /* reset tzabbrevcache, which may contain results from old table */ + memset(tzabbrevcache, 0, sizeof(tzabbrevcache)); } /* diff --git a/src/include/utils/datetime.h b/src/include/utils/datetime.h index 7fe12a4ea7..53a1c69eda 100644 --- a/src/include/utils/datetime.h +++ b/src/include/utils/datetime.h @@ -351,6 +351,8 @@ extern pg_tz *DecodeTimezoneNameToTz(const char *tzname); extern int DecodeTimezoneAbbrevPrefix(const char *str, int *offset, pg_tz **tz); +extern void ClearTimeZoneAbbrevCache(void); + extern int j2day(int date); extern struct Node *TemporalSimplify(int32 max_precis, struct Node *node); diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 9f83ecf181..e11f8e781b 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3030,6 +3030,7 @@ TypeCat TypeFuncClass TypeInfo TypeName +TzAbbrevCache U32 U8 UChar -- 2.43.5