Re: Pluggable storage

On Sat, Sep 9, 2017 at 1:23 PM, Haribabu Kommi <kommi(dot)haribabu(at)gmail(dot)com>
wrote:

>
> I rebased the patch to the latest master and also fixed the duplicate OID
> and some slot fixes. Updated patches are attached.
>

While analyzing the removal of HeapScanDesc usage other than heap
modules, The mostly used member is "*rs_cbuf*" for the purpose of locking
the buffer, visibility checks and marking it as dirty. The buffer is
tightly integrated
with the visibility. Buffer may not be required for some storage routines
where
the data is always in the memory and etc.

I found the references of its structure members in the following
places. I feel other than the following 4 parameters, rest of them needs to
be
moved into their corresponding storage routines.

* Relation rs_rd; /* heap relation descriptor */*
* Snapshot rs_snapshot; /* snapshot to see */*
* int rs_nkeys; /* number of scan keys */*
* ScanKey rs_key; /* array of scan key descriptors */*

But currently I am treating the "*rs_cbuf" *also a needed member and also
expecting all storage routines will be provide it. Or we may need a another
approach to mark the buffer as dirty.

Suggestions?

Following are the rest of the parameters that are used
outside the heap.

* BlockNumber rs_nblocks; /* total number of blocks in rel */*

pgstattuple.c, tsm_system_rows.c, tsm_system_time.c, system.c
nodeBitmapheapscan.c nodesamplescan.c,
*Mostly for the purpose of checking the number of blocks in a rel.*

* BufferAccessStrategy rs_strategy; /* access strategy for reads */*

pgstattuple.c

* bool rs_pageatatime; /* verify visibility page-at-a-time? */*
* BlockNumber rs_startblock; /* block # to start at */*
* bool rs_syncscan; /* report location to syncscan logic? */*
* bool rs_inited; /* false = scan not init'd yet */*

nodesamplescan.c

* HeapTupleData rs_ctup; /* current tuple in scan, if any */*

*genam.c, nodeBitmapHeapscan.c, nodesamplescan.c*
*Used for retrieving the last scanned tuple.*

* BlockNumber rs_cblock; /* current block # in scan, if any */*

*index.c, nodesamplescan.c*

* Buffer rs_cbuf; /* current buffer in scan, if any */*

*pgrowlocks.c, pgstattuple.c, genam.c, index.c, cluster.c,*
*tablecmds.c, nodeBitmapHeapscan.c, nodesamplescan.c*
*Mostly used for Locking the Buffer.*

* ParallelHeapScanDesc rs_parallel; /* parallel scan information */*

*nodeseqscan.c*

* int rs_cindex; /* current tuple's index in vistuples */*

*nodeBitmapHeapScan.c*

* int rs_ntuples; /* number of visible tuples on page */*
* OffsetNumber rs_vistuples[MaxHeapTuplesPerPage]; /* their offsets */*

*tsm_system_rows.c, nodeBitmapHeapscan.c, nodesamplescan.c*
*Used for retrieve the offsets mainly Bitmap and sample scans.*

I think rest of the above parameters usage other than heap can be changed
once the Bitmap and Sample scans are modified to use the storage routines
while returning the tuple instead of their own implementations. I feel these
scans are the major users of the rest of the parameters. This approach may
need to some more API's to get rid of Bitmap and sample scan's own
implementation.

suggestions?

Regards,
Hari Babu
Fujitsu Australia