Allow ReadStream to be consumed as raw block numbers.
authorThomas Munro <[email protected]>
Tue, 17 Sep 2024 21:20:59 +0000 (09:20 +1200)
committerThomas Munro <[email protected]>
Tue, 17 Sep 2024 23:34:28 +0000 (11:34 +1200)
Commits 041b9680 and 6377e12a changed the interface of
scan_analyze_next_block() to take a ReadStream instead of a BlockNumber
and a BufferAccessStrategy, and to return a value to indicate when the
stream has run out of blocks.

This caused integration problems for at least one known extension that
uses specially encoded BlockNumber values that map to different
underlying storage, because acquire_sample_rows() sets up the stream so
that read_stream_next_buffer() reads blocks from the main fork of the
relation's SMgrRelation.

Provide read_stream_next_block(), as a way for such an extension to
access the stream of raw BlockNumbers directly and forward them to its
own ReadBuffer() calls after decoding, as it could in earlier releases.
The new function returns the BlockNumber and BufferAccessStrategy that
were previously passed directly to scan_analyze_next_block().
Alternatively, an extension could wrap the stream of BlockNumbers in
another ReadStream with a callback that performs any decoding required
to arrive at real storage manager BlockNumber values, so that it could
benefit from the I/O combining and concurrency provided by
read_stream.c.

Another class of table access method that does nothing in
scan_analyze_next_block() because it is not block-oriented could use
this function to control the number of block sampling loops.  It could
match the previous behavior with "return read_stream_next_block(stream,
&bas) != InvalidBlockNumber".

Ongoing work is expected to provide better ANALYZE support for table
access methods that don't behave like heapam with respect to storage
blocks, but that will be for future releases.

Back-patch to 17.

Reported-by: Mats Kindahl <[email protected]>
Reviewed-by: Mats Kindahl <[email protected]>
Discussion: https://postgr.es/m/CA%2B14425%2BCcm07ocG97Fp%2BFrD9xUXqmBKFvecp0p%2BgV2YYR258Q%40mail.gmail.com

src/backend/storage/aio/read_stream.c
src/include/storage/read_stream.h

index 7f0e07d9586a434c77054db2920f3f5c9ca22d9e..b8ea1180e25990d6f0263ba065acc9e7c402d06e 100644 (file)
@@ -797,6 +797,20 @@ read_stream_next_buffer(ReadStream *stream, void **per_buffer_data)
    return buffer;
 }
 
+/*
+ * Transitional support for code that would like to perform or skip reads
+ * itself, without using the stream.  Returns, and consumes, the next block
+ * number that would be read by the stream's look-ahead algorithm, or
+ * InvalidBlockNumber if the end of the stream is reached.  Also reports the
+ * strategy that would be used to read it.
+ */
+BlockNumber
+read_stream_next_block(ReadStream *stream, BufferAccessStrategy *strategy)
+{
+   *strategy = stream->ios[0].op.strategy;
+   return read_stream_get_block(stream, NULL);
+}
+
 /*
  * Reset a read stream by releasing any queued up buffers, allowing the stream
  * to be used again for different blocks.  This can be used to clear an
index 2f94ee744b951c3778eb23955bf89efc0e82b2bf..2f98b600f7aae75bb12727ecedbcc4e2af886002 100644 (file)
@@ -68,6 +68,8 @@ extern ReadStream *read_stream_begin_relation(int flags,
                                              void *callback_private_data,
                                              size_t per_buffer_data_size);
 extern Buffer read_stream_next_buffer(ReadStream *stream, void **per_buffer_data);
+extern BlockNumber read_stream_next_block(ReadStream *stream,
+                                         BufferAccessStrategy *strategy);
 extern ReadStream *read_stream_begin_smgr_relation(int flags,
                                                   BufferAccessStrategy strategy,
                                                   SMgrRelation smgr,