Hi,
Please find attached the next version of the extensible toast
support. There basically are two changes:
* handle indirect toast tuples properly in heap_tuple_fetch_attr
and related places
* minor comment adjustments
Greetings,
Andres Freund
--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services
>From 5c7079067f870a65d65ee09cccbf743c88b64c82 Mon Sep 17 00:00:00 2001
From: Andres Freund <and...@anarazel.de>
Date: Tue, 11 Jun 2013 23:25:26 +0200
Subject: [PATCH] Add support for multiple kinds of external toast datums
There are several usecases where our current representation of external toast
datums is limiting:
* adding new compression schemes
* avoidance of repeated detoasting
* externally decoded toast tuples
For that support 'tags' on external (varattrib_1b_e) varlenas which recoin the
current va_len_1be field to store the tag (or type) of a varlena. To determine
the actual length a macro VARTAG_SIZE(tag) is added which can be used to map
from a tag to the actual length.
This patch adds support for 'indirect' tuples which point to some externally
allocated memory containing a toast tuple. It also implements the stub for a
different compression algorithm.
---
src/backend/access/heap/tuptoaster.c | 142 +++++++++++++++++++++++++++++++----
src/include/c.h | 2 +
src/include/postgres.h | 84 +++++++++++++++------
3 files changed, 191 insertions(+), 37 deletions(-)
diff --git a/src/backend/access/heap/tuptoaster.c b/src/backend/access/heap/tuptoaster.c
index fc37ceb..5ee5e33 100644
--- a/src/backend/access/heap/tuptoaster.c
+++ b/src/backend/access/heap/tuptoaster.c
@@ -87,11 +87,11 @@ static struct varlena *toast_fetch_datum_slice(struct varlena * attr,
* heap_tuple_fetch_attr -
*
* Public entry point to get back a toasted value from
- * external storage (possibly still in compressed format).
+ * external source (possibly still in compressed format).
*
* This will return a datum that contains all the data internally, ie, not
- * relying on external storage, but it can still be compressed or have a short
- * header.
+ * relying on external storage or memory, but it can still be compressed or
+ * have a short header.
----------
*/
struct varlena *
@@ -99,13 +99,35 @@ heap_tuple_fetch_attr(struct varlena * attr)
{
struct varlena *result;
- if (VARATT_IS_EXTERNAL(attr))
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
{
/*
* This is an external stored plain value
*/
result = toast_fetch_datum(attr);
}
+ else if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ {
+ /*
+ * copy into the caller's memory context. That's not required in all
+ * cases but sufficient for now since this is mainly used when we need
+ * to persist a Datum for unusually long time, like in a HOLD cursor.
+ */
+ struct varatt_indirect redirect;
+ VARATT_EXTERNAL_GET_POINTER(redirect, attr);
+ attr = (struct varlena *)redirect.pointer;
+
+ /* nested indirect Datums aren't allowed */
+ Assert(!VARATT_IS_EXTERNAL_INDIRECT(attr));
+
+ /* doesn't make much sense, but better handle it */
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
+ return heap_tuple_fetch_attr(attr);
+
+ /* copy datum verbatim */
+ result = (struct varlena *) palloc(VARSIZE_ANY(attr));
+ memcpy(result, attr, VARSIZE_ANY(attr));
+ }
else
{
/*
@@ -128,7 +150,7 @@ heap_tuple_fetch_attr(struct varlena * attr)
struct varlena *
heap_tuple_untoast_attr(struct varlena * attr)
{
- if (VARATT_IS_EXTERNAL(attr))
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
{
/*
* This is an externally stored datum --- fetch it back from there
@@ -145,6 +167,17 @@ heap_tuple_untoast_attr(struct varlena * attr)
pfree(tmp);
}
}
+ else if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ {
+ struct varatt_indirect redirect;
+ VARATT_EXTERNAL_GET_POINTER(redirect, attr);
+ attr = (struct varlena *)redirect.pointer;
+
+ /* nested indirect Datums aren't allowed */
+ Assert(!VARATT_IS_EXTERNAL_INDIRECT(attr));
+
+ attr = heap_tuple_untoast_attr(attr);
+ }
else if (VARATT_IS_COMPRESSED(attr))
{
/*
@@ -191,7 +224,7 @@ heap_tuple_untoast_attr_slice(struct varlena * attr,
char *attrdata;
int32 attrsize;
- if (VARATT_IS_EXTERNAL(attr))
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
{
struct varatt_external toast_pointer;
@@ -204,6 +237,17 @@ heap_tuple_untoast_attr_slice(struct varlena * attr,
/* fetch it back (compressed marker will get set automatically) */
preslice = toast_fetch_datum(attr);
}
+ else if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ {
+ struct varatt_indirect redirect;
+ VARATT_EXTERNAL_GET_POINTER(redirect, attr);
+
+ /* nested indirect Datums aren't allowed */
+ Assert(!VARATT_IS_EXTERNAL_INDIRECT(redirect.pointer));
+
+ return heap_tuple_untoast_attr_slice(redirect.pointer,
+ sliceoffset, slicelength);
+ }
else
preslice = attr;
@@ -267,7 +311,7 @@ toast_raw_datum_size(Datum value)
struct varlena *attr = (struct varlena *) DatumGetPointer(value);
Size result;
- if (VARATT_IS_EXTERNAL(attr))
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
{
/* va_rawsize is the size of the original datum -- including header */
struct varatt_external toast_pointer;
@@ -275,6 +319,16 @@ toast_raw_datum_size(Datum value)
VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
result = toast_pointer.va_rawsize;
}
+ else if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ {
+ struct varatt_indirect toast_pointer;
+ VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
+
+ /* nested indirect Datums aren't allowed */
+ Assert(!VARATT_IS_EXTERNAL_INDIRECT(toast_pointer.pointer));
+
+ return toast_raw_datum_size(PointerGetDatum(toast_pointer.pointer));
+ }
else if (VARATT_IS_COMPRESSED(attr))
{
/* here, va_rawsize is just the payload size */
@@ -308,7 +362,7 @@ toast_datum_size(Datum value)
struct varlena *attr = (struct varlena *) DatumGetPointer(value);
Size result;
- if (VARATT_IS_EXTERNAL(attr))
+ if (VARATT_IS_EXTERNAL_ONDISK(attr))
{
/*
* Attribute is stored externally - return the extsize whether
@@ -320,6 +374,16 @@ toast_datum_size(Datum value)
VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
result = toast_pointer.va_extsize;
}
+ else if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ {
+ struct varatt_indirect toast_pointer;
+ VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
+
+ /* nested indirect Datums aren't allowed */
+ Assert(!VARATT_IS_EXTERNAL_INDIRECT(attr));
+
+ return toast_datum_size(PointerGetDatum(toast_pointer.pointer));
+ }
else if (VARATT_IS_SHORT(attr))
{
result = VARSIZE_SHORT(attr);
@@ -387,12 +451,56 @@ toast_delete(Relation rel, HeapTuple oldtup)
{
Datum value = toast_values[i];
- if (!toast_isnull[i] && VARATT_IS_EXTERNAL(PointerGetDatum(value)))
+ if (toast_isnull[i])
+ continue;
+ else if (VARATT_IS_EXTERNAL_ONDISK(PointerGetDatum(value)))
toast_delete_datum(rel, value);
+ else if (VARATT_IS_EXTERNAL_INDIRECT(PointerGetDatum(value)))
+ elog(ERROR, "deleteing a tuple indirect datums doesn't make sense");
}
}
}
+/* ----------
+ * toast_datum_differs -
+ *
+ * Determine whether two toasted Datums are the same and don't have to be
+ * stored again.
+ * ----------
+ */
+static bool
+toast_datum_differs(struct varlena *old_value, struct varlena *new_value)
+{
+ Assert(VARATT_IS_EXTERNAL(old_value));
+ Assert(VARATT_IS_EXTERNAL(new_value));
+
+ /* fast path for the common case where we have the toast oid available */
+ if (VARATT_IS_EXTERNAL_ONDISK(old_value) &&
+ VARATT_IS_EXTERNAL_ONDISK(new_value))
+ return memcmp((char *) old_value, (char *) new_value,
+ VARSIZE_EXTERNAL(old_value)) != 0;
+
+ /*
+ * compare size of tuples, so we don't uselessly detoast/decompress tuples
+ * if they can't be the same anyway.
+ */
+ if (toast_raw_datum_size(PointerGetDatum(old_value)) !=
+ toast_raw_datum_size(PointerGetDatum(new_value)))
+ return false;
+
+ old_value = heap_tuple_untoast_attr(old_value);
+ new_value = heap_tuple_untoast_attr(new_value);
+
+ Assert(!VARATT_IS_EXTERNAL(old_value));
+ Assert(!VARATT_IS_EXTERNAL(new_value));
+ Assert(!VARATT_IS_COMPRESSED(old_value));
+ Assert(!VARATT_IS_COMPRESSED(new_value));
+ Assert(VARSIZE_ANY_EXHDR(old_value) == VARSIZE_ANY_EXHDR(new_value));
+
+ /* compare payload, we're fine with unaligned data */
+ return memcmp(VARDATA_ANY(old_value), VARDATA_ANY(new_value),
+ VARSIZE_ANY_EXHDR(old_value)) != 0;
+}
/* ----------
* toast_insert_or_update -
@@ -497,8 +605,7 @@ toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup,
VARATT_IS_EXTERNAL(old_value))
{
if (toast_isnull[i] || !VARATT_IS_EXTERNAL(new_value) ||
- memcmp((char *) old_value, (char *) new_value,
- VARSIZE_EXTERNAL(old_value)) != 0)
+ toast_datum_differs(old_value, new_value))
{
/*
* The old external stored value isn't needed any more
@@ -1258,6 +1365,8 @@ toast_save_datum(Relation rel, Datum value,
int32 data_todo;
Pointer dval = DatumGetPointer(value);
+ Assert(!VARATT_IS_EXTERNAL(value));
+
/*
* Open the toast relation and its index. We can use the index to check
* uniqueness of the OID we assign to the toasted item, even though it has
@@ -1341,7 +1450,7 @@ toast_save_datum(Relation rel, Datum value,
{
struct varatt_external old_toast_pointer;
- Assert(VARATT_IS_EXTERNAL(oldexternal));
+ Assert(VARATT_IS_EXTERNAL_ONDISK(oldexternal));
/* Must copy to access aligned fields */
VARATT_EXTERNAL_GET_POINTER(old_toast_pointer, oldexternal);
if (old_toast_pointer.va_toastrelid == rel->rd_toastoid)
@@ -1456,7 +1565,7 @@ toast_save_datum(Relation rel, Datum value,
* Create the TOAST pointer value that we'll return
*/
result = (struct varlena *) palloc(TOAST_POINTER_SIZE);
- SET_VARSIZE_EXTERNAL(result, TOAST_POINTER_SIZE);
+ SET_VARTAG_EXTERNAL(result, VARTAG_ONDISK);
memcpy(VARDATA_EXTERNAL(result), &toast_pointer, sizeof(toast_pointer));
return PointerGetDatum(result);
@@ -1480,7 +1589,7 @@ toast_delete_datum(Relation rel, Datum value)
SysScanDesc toastscan;
HeapTuple toasttup;
- if (!VARATT_IS_EXTERNAL(attr))
+ if (!VARATT_IS_EXTERNAL_ONDISK(attr))
return;
/* Must copy to access aligned fields */
@@ -1608,6 +1717,9 @@ toast_fetch_datum(struct varlena * attr)
char *chunkdata;
int32 chunksize;
+ if (VARATT_IS_EXTERNAL_INDIRECT(attr))
+ elog(ERROR, "shouldn't be called for indirect tuples");
+
/* Must copy to access aligned fields */
VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
@@ -1775,7 +1887,7 @@ toast_fetch_datum_slice(struct varlena * attr, int32 sliceoffset, int32 length)
int32 chcpystrt;
int32 chcpyend;
- Assert(VARATT_IS_EXTERNAL(attr));
+ Assert(VARATT_IS_EXTERNAL_ONDISK(attr));
/* Must copy to access aligned fields */
VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
diff --git a/src/include/c.h b/src/include/c.h
index f2c9e12..7193af6 100644
--- a/src/include/c.h
+++ b/src/include/c.h
@@ -573,6 +573,8 @@ typedef NameData *Name;
#define AssertMacro(condition) ((void)true)
#define AssertArg(condition)
#define AssertState(condition)
+#define TrapMacro(condition, errorType) (true)
+
#elif defined(FRONTEND)
#include <assert.h>
diff --git a/src/include/postgres.h b/src/include/postgres.h
index 30e1dee..d57569f 100644
--- a/src/include/postgres.h
+++ b/src/include/postgres.h
@@ -54,23 +54,53 @@
*/
/*
- * struct varatt_external is a "TOAST pointer", that is, the information
- * needed to fetch a stored-out-of-line Datum. The data is compressed
- * if and only if va_extsize < va_rawsize - VARHDRSZ. This struct must not
- * contain any padding, because we sometimes compare pointers using memcmp.
+ * struct varatt_external is a "TOAST pointer", that is, the information needed
+ * to fetch a Datum stored in an out-of-line on-disk Datum. The data is
+ * compressed if and only if va_extsize < va_rawsize - VARHDRSZ. This struct
+ * must not contain any padding, because we sometimes compare pointers using
+ * memcmp.
*
* Note that this information is stored unaligned within actual tuples, so
* you need to memcpy from the tuple into a local struct variable before
* you can look at these fields! (The reason we use memcmp is to avoid
* having to do that just to detect equality of two TOAST pointers...)
*/
-struct varatt_external
+typedef struct varatt_external
{
int32 va_rawsize; /* Original data size (includes header) */
int32 va_extsize; /* External saved size (doesn't) */
Oid va_valueid; /* Unique ID of value within TOAST table */
Oid va_toastrelid; /* RelID of TOAST table containing it */
-};
+} varatt_external;
+
+/*
+ * Out-of-line Datum thats stored in memory in contrast to varatt_external
+ * pointers which points to data in an external toast relation.
+ *
+ * Note that just as varatt_external's the redirection itsel is stored
+ * unaligned within the tuple.
+ */
+typedef struct varatt_indirect
+{
+ struct varlena *pointer; /* Pointer to in-memory varlena */
+} varatt_indirect;
+
+
+/*
+ * Type of external toast datum stored. The peculiar value for VARTAG_ONDISK
+ * comes from the requirement for on-disk compatibility with the older
+ * definitions of varattrib_1b_e where v_tag was named va_len_1be...
+ */
+typedef enum vartag_external
+{
+ VARTAG_INDIRECT = 1,
+ VARTAG_ONDISK = 18
+} vartag_external;
+
+#define VARTAG_SIZE(tag) \
+ ((tag) == VARTAG_INDIRECT ? sizeof(varatt_indirect) : \
+ (tag) == VARTAG_ONDISK ? sizeof(varatt_external) : \
+ TrapMacro(false, "unknown vartag"))
/*
* These structs describe the header of a varlena object that may have been
@@ -102,11 +132,12 @@ typedef struct
char va_data[1]; /* Data begins here */
} varattrib_1b;
+/* inline portion of a short varlena pointing to an external resource */
typedef struct
{
uint8 va_header; /* Always 0x80 or 0x01 */
- uint8 va_len_1be; /* Physical length of datum */
- char va_data[1]; /* Data (for now always a TOAST pointer) */
+ uint8 va_tag; /* Type of datum */
+ char va_data[1]; /* Data (of the type indicated by va_tag) */
} varattrib_1b_e;
/*
@@ -130,6 +161,9 @@ typedef struct
* first byte. Also, it is not possible for a 1-byte length word to be zero;
* this lets us disambiguate alignment padding bytes from the start of an
* unaligned datum. (We now *require* pad bytes to be filled with zero!)
+ *
+ * In TOAST datums the tag field in varattrib_1b_e is used to discern whether
+ * its an indirection pointer or more commonly an on-disk tuple.
*/
/*
@@ -161,8 +195,8 @@ typedef struct
(((varattrib_4b *) (PTR))->va_4byte.va_header & 0x3FFFFFFF)
#define VARSIZE_1B(PTR) \
(((varattrib_1b *) (PTR))->va_header & 0x7F)
-#define VARSIZE_1B_E(PTR) \
- (((varattrib_1b_e *) (PTR))->va_len_1be)
+#define VARTAG_1B_E(PTR) \
+ (((varattrib_1b_e *) (PTR))->va_tag)
#define SET_VARSIZE_4B(PTR,len) \
(((varattrib_4b *) (PTR))->va_4byte.va_header = (len) & 0x3FFFFFFF)
@@ -170,9 +204,9 @@ typedef struct
(((varattrib_4b *) (PTR))->va_4byte.va_header = ((len) & 0x3FFFFFFF) | 0x40000000)
#define SET_VARSIZE_1B(PTR,len) \
(((varattrib_1b *) (PTR))->va_header = (len) | 0x80)
-#define SET_VARSIZE_1B_E(PTR,len) \
+#define SET_VARTAG_1B_E(PTR,tag) \
(((varattrib_1b_e *) (PTR))->va_header = 0x80, \
- ((varattrib_1b_e *) (PTR))->va_len_1be = (len))
+ ((varattrib_1b_e *) (PTR))->va_tag = (tag))
#else /* !WORDS_BIGENDIAN */
#define VARATT_IS_4B(PTR) \
@@ -193,8 +227,8 @@ typedef struct
((((varattrib_4b *) (PTR))->va_4byte.va_header >> 2) & 0x3FFFFFFF)
#define VARSIZE_1B(PTR) \
((((varattrib_1b *) (PTR))->va_header >> 1) & 0x7F)
-#define VARSIZE_1B_E(PTR) \
- (((varattrib_1b_e *) (PTR))->va_len_1be)
+#define VARTAG_1B_E(PTR) \
+ (((varattrib_1b_e *) (PTR))->va_tag)
#define SET_VARSIZE_4B(PTR,len) \
(((varattrib_4b *) (PTR))->va_4byte.va_header = (((uint32) (len)) << 2))
@@ -202,12 +236,12 @@ typedef struct
(((varattrib_4b *) (PTR))->va_4byte.va_header = (((uint32) (len)) << 2) | 0x02)
#define SET_VARSIZE_1B(PTR,len) \
(((varattrib_1b *) (PTR))->va_header = (((uint8) (len)) << 1) | 0x01)
-#define SET_VARSIZE_1B_E(PTR,len) \
+#define SET_VARTAG_1B_E(PTR,tag) \
(((varattrib_1b_e *) (PTR))->va_header = 0x01, \
- ((varattrib_1b_e *) (PTR))->va_len_1be = (len))
+ ((varattrib_1b_e *) (PTR))->va_tag = (tag))
#endif /* WORDS_BIGENDIAN */
-#define VARHDRSZ_SHORT 1
+#define VARHDRSZ_SHORT offsetof(varattrib_1b, va_data)
#define VARATT_SHORT_MAX 0x7F
#define VARATT_CAN_MAKE_SHORT(PTR) \
(VARATT_IS_4B_U(PTR) && \
@@ -215,7 +249,7 @@ typedef struct
#define VARATT_CONVERTED_SHORT_SIZE(PTR) \
(VARSIZE(PTR) - VARHDRSZ + VARHDRSZ_SHORT)
-#define VARHDRSZ_EXTERNAL 2
+#define VARHDRSZ_EXTERNAL offsetof(varattrib_1b_e, va_data)
#define VARDATA_4B(PTR) (((varattrib_4b *) (PTR))->va_4byte.va_data)
#define VARDATA_4B_C(PTR) (((varattrib_4b *) (PTR))->va_compressed.va_data)
@@ -249,26 +283,32 @@ typedef struct
#define VARSIZE_SHORT(PTR) VARSIZE_1B(PTR)
#define VARDATA_SHORT(PTR) VARDATA_1B(PTR)
-#define VARSIZE_EXTERNAL(PTR) VARSIZE_1B_E(PTR)
+#define VARTAG_EXTERNAL(PTR) VARTAG_1B_E(PTR)
+#define VARSIZE_EXTERNAL(PTR) (VARHDRSZ_EXTERNAL + VARTAG_SIZE(VARTAG_EXTERNAL(PTR)))
#define VARDATA_EXTERNAL(PTR) VARDATA_1B_E(PTR)
#define VARATT_IS_COMPRESSED(PTR) VARATT_IS_4B_C(PTR)
#define VARATT_IS_EXTERNAL(PTR) VARATT_IS_1B_E(PTR)
+#define VARATT_IS_EXTERNAL_ONDISK(PTR) \
+ (VARATT_IS_EXTERNAL(PTR) && VARTAG_EXTERNAL(PTR) == VARTAG_ONDISK)
+#define VARATT_IS_EXTERNAL_INDIRECT(PTR) \
+ (VARATT_IS_EXTERNAL(PTR) && VARTAG_EXTERNAL(PTR) == VARTAG_INDIRECT)
#define VARATT_IS_SHORT(PTR) VARATT_IS_1B(PTR)
#define VARATT_IS_EXTENDED(PTR) (!VARATT_IS_4B_U(PTR))
#define SET_VARSIZE(PTR, len) SET_VARSIZE_4B(PTR, len)
#define SET_VARSIZE_SHORT(PTR, len) SET_VARSIZE_1B(PTR, len)
#define SET_VARSIZE_COMPRESSED(PTR, len) SET_VARSIZE_4B_C(PTR, len)
-#define SET_VARSIZE_EXTERNAL(PTR, len) SET_VARSIZE_1B_E(PTR, len)
+
+#define SET_VARTAG_EXTERNAL(PTR, tag) SET_VARTAG_1B_E(PTR, tag)
#define VARSIZE_ANY(PTR) \
- (VARATT_IS_1B_E(PTR) ? VARSIZE_1B_E(PTR) : \
+ (VARATT_IS_1B_E(PTR) ? VARSIZE_EXTERNAL(PTR) : \
(VARATT_IS_1B(PTR) ? VARSIZE_1B(PTR) : \
VARSIZE_4B(PTR)))
#define VARSIZE_ANY_EXHDR(PTR) \
- (VARATT_IS_1B_E(PTR) ? VARSIZE_1B_E(PTR)-VARHDRSZ_EXTERNAL : \
+ (VARATT_IS_1B_E(PTR) ? VARSIZE_EXTERNAL(PTR)-VARHDRSZ_EXTERNAL : \
(VARATT_IS_1B(PTR) ? VARSIZE_1B(PTR)-VARHDRSZ_SHORT : \
VARSIZE_4B(PTR)-VARHDRSZ))
--
1.8.3.1
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
