netcdf-c/libnczarr/zchunking.c
Dennis Heimbigner df3636b959 Mitigate S3 test interference + Unlimited Dimensions in NCZarr
This PR started as an attempt to add unlimited dimensions to NCZarr.
It did that, but this exposed significant problems with test interference.
So this PR is mostly about fixing -- well mitigating anyway -- test
interference.

The problem of test interference is now documented in the document docs/internal.md.
The solutions implemented here are also describe in that document.
The solution is somewhat fragile but multiple cleanup mechanisms
are provided. Note that this feature requires that the
AWS command line utility must be installed.

## Unlimited Dimensions.
The existing NCZarr extensions to Zarr are modified to support unlimited dimensions.
NCzarr extends the Zarr meta-data for the ".zgroup" object to include netcdf-4 model extensions. This information is stored in ".zgroup" as dictionary named "_nczarr_group".
Inside "_nczarr_group", there is a key named "dims" that stores information about netcdf-4 named dimensions. The value of "dims" is a dictionary whose keys are the named dimensions. The value associated with each dimension name has one of two forms
Form 1 is a special case of form 2, and is kept for backward compatibility. Whenever a new file is written, it uses format 1 if possible, otherwise format 2.
* Form 1: An integer representing the size of the dimension, which is used for simple named dimensions.
* Form 2: A dictionary with the following keys and values"
   - "size" with an integer value representing the (current) size of the dimension.
   - "unlimited" with a value of either "1" or "0" to indicate if this dimension is an unlimited dimension.

For Unlimited dimensions, the size is initially zero, and as variables extend the length of that dimension, the size value for the dimension increases.
That dimension size is shared by all arrays referencing that dimension, so if one array extends an unlimited dimension, it is implicitly extended for all other arrays that reference that dimension.
This is the standard semantics for unlimited dimensions.

Adding unlimited dimensions required a number of other changes to the NCZarr code-base. These included the following.
* Did a partial refactor of the slice handling code in zwalk.c to clean it up.
* Added a number of tests for unlimited dimensions derived from the same test in nc_test4.
* Added several NCZarr specific unlimited tests; more are needed.
* Add test of endianness.

## Misc. Other Changes
* Modify libdispatch/ncs3sdk_aws.cpp to optionally support use of the
   AWS Transfer Utility mechanism. This is controlled by the
   ```#define TRANSFER```` command in that file. It defaults to being disabled.
* Parameterize both the standard Unidata S3 bucket (S3TESTBUCKET) and the netcdf-c test data prefix (S3TESTSUBTREE).
* Fixed an obscure memory leak in ncdump.
* Removed some obsolete unit testing code and test cases.
* Uncovered a bug in the netcdf-c handling of big-endian floats and doubles. Have not fixed yet. See tst_h5_endians.c.
* Renamed some nczarr_tests testcases to avoid name conflicts with nc_test4.
* Modify the semantics of zmap\#ncsmap_write to only allow total rewrite of objects.
* Modify the semantics of zodom to properly handle stride > 1.
* Add a truncate operation to the libnczarr zmap code.
2023-09-26 16:56:48 -06:00

315 lines
10 KiB
C

/*********************************************************************
* Copyright 2018, UCAR/Unidata
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
*********************************************************************/
#include "zincludes.h"
#define MAX(a,b) ((a)>(b)?(a):(b))
#define MIN(a,b) ((a)<(b)?(a):(b))
static int pcounter = 0;
/* Forward */
static int compute_intersection(const NCZSlice* slice, const size64_t chunklen, unsigned char isunlimited, NCZChunkRange* range);
static void skipchunk(const NCZSlice* slice, NCZProjection* projection);
static int verifyslice(const NCZSlice* slice);
/**************************************************/
/* Goal:create a vector of chunk ranges: one for each slice in
the top-level input. For each slice, compute the index (not
absolute position) of the first chunk that intersects the slice
and the index of the last chunk that intersects the slice.
In practice, the count = last - first + 1 is stored instead of the last index.
Note that this n-dim array of indices may have holes in it if the slice stride
is greater than the chunk length.
@param rank variable rank
@param slices the complete set of slices |slices| == R
@param ncr (out) the vector of computed chunk ranges.
@return NC_EXXX error code
*/
int
NCZ_compute_chunk_ranges(
struct Common* common,
const NCZSlice* slices, /* the complete set of slices |slices| == R*/
NCZChunkRange* ncr)
{
int stat = NC_NOERR;
int i;
int rank = common->rank;
for(i=0;i<rank;i++) {
if((stat = compute_intersection(&slices[i],common->chunklens[i],common->isunlimited[i],&ncr[i])))
goto done;
}
done:
return stat;
}
/**
@param Compute chunk range for a single slice.
@param chunklen size of the chunk
@param isunlimited if corresponding dim is unlimited
@param range (out) the range of chunks covered by this slice
@return NC_EXX error code
*/
static int
compute_intersection(
const NCZSlice* slice,
size64_t chunklen,
unsigned char isunlimited,
NCZChunkRange* range)
{
range->start = floordiv(slice->start, chunklen);
range->stop = ceildiv(slice->stop, chunklen);
return NC_NOERR;
}
/**
Compute the projection of a slice as applied to n'th chunk.
A projection defines the set of grid points touched within a
chunk by a slice. This set of points is the "projection"
of the slice onto the chunk.
This is somewhat complex because:
1. for the first projection, the start is the slice start,
but after that, we have to take into account that for
a non-one stride, the start point in a projection may
be offset by some value in the range of 0..(slice.stride-1).
2. The stride might be so large as to completely skip some chunks.
@return NC_NOERR if ok
@return NC_ERANGE if chunk skipped
@return NC_EXXXX if failed
*/
int
NCZ_compute_projections(struct Common* common, int r, size64_t chunkindex, const NCZSlice* slice, size_t n, NCZProjection* projections)
{
int stat = NC_NOERR;
NCZProjection* projection = NULL;
NCZProjection* prev = NULL;
size64_t dimlen = common->dimlens[r]; /* the dimension length for r'th dimension */
size64_t chunklen = common->chunklens[r]; /* the chunk length corresponding to the dimension */
size64_t abslimit;
projection = &projections[n];
if(n > 0) {
/* Find last non-skipped projection */
int i;
for(i=n-1;i>=0;i--) { /* walk backward */
if(!projections[i].skip) {
prev = &projections[i];
break;
}
}
if(prev == NULL) {stat = NC_ENCZARR; goto done;}
}
projection->id = ++pcounter;
projection->chunkindex = chunkindex;
projection->offset = chunklen * chunkindex; /* with respect to dimension (WRD) */
/* limit in the n'th touched chunk, taking dimlen and stride->stop into account. */
abslimit = (chunkindex + 1) * chunklen;
if(abslimit > slice->stop) abslimit = slice->stop;
if(abslimit > dimlen) abslimit = dimlen;
projection->limit = abslimit - projection->offset;
/* See if the next point after the last one in prev lands in the current projection.
If not, then we have skipped the current chunk. Also take limit into account.
Note by definition, n must be greater than zero because we always start in a relevant chunk.
*/
if(n == 0) {
/*initial case: original slice start is in 1st projection */
projection->first = slice->start - projection->offset;
projection->iopos = 0;
} else { /* n > 0 */
/* Use absolute offsets for these computations to avoid negative values */
size64_t abslastpoint, absnextpoint, absthislast;
/* abs last point touched in prev projection */
abslastpoint = prev->offset + prev->last;
/* Compute the abs last touchable point in this chunk */
absthislast = projection->offset + projection->limit;
/* Compute next point touched after the last point touched in previous projection;
note that the previous projection might be wrt a chunk other than the immediately preceding
one (because the intermediate ones were skipped).
*/
absnextpoint = abslastpoint + slice->stride; /* abs next point to be touched */
if(absnextpoint >= absthislast) { /* this chunk is being skipped */
skipchunk(slice,projection);
goto done;
}
/* Compute start point in this chunk */
/* basically absnextpoint - abs start of this projection */
projection->first = absnextpoint - projection->offset;
/* Compute the memory location of this first point in this chunk */
projection->iopos = ceildiv((projection->offset - slice->start),slice->stride);
}
if(slice->stop > abslimit)
projection->stop = chunklen;
else
projection->stop = slice->stop - projection->offset;
projection->iocount = ceildiv((projection->stop - projection->first),slice->stride);
/* Compute the slice relative to this chunk.
Recall the possibility that start+stride >= projection->limit */
projection->chunkslice.start = projection->first;
projection->chunkslice.stop = projection->stop;
projection->chunkslice.stride = slice->stride;
projection->chunkslice.len = chunklen;
/* Last place to be touched */
projection->last = projection->first + (slice->stride * (projection->iocount - 1));
projection->memslice.start = projection->iopos;
projection->memslice.stop = projection->iopos + projection->iocount;
projection->memslice.stride = 1;
// projection->memslice.stride = slice->stride;
// projection->memslice.len = projection->memslice.stop;
projection->memslice.len = common->memshape[r];
#ifdef NEVERUSE
projection->memslice.len = dimlen;
projection->memslice.len = chunklen;
#endif
if(!verifyslice(&projection->memslice) || !verifyslice(&projection->chunkslice))
{stat = NC_ECONSTRAINT; goto done;}
done:
return stat;
}
static void
skipchunk(const NCZSlice* slice, NCZProjection* projection)
{
projection->skip = 1;
projection->first = 0;
projection->last = 0;
projection->iopos = ceildiv(projection->offset - slice->start, slice->stride);
projection->iocount = 0;
projection->chunkslice.start = 0;
projection->chunkslice.stop = 0;
projection->chunkslice.stride = 1;
projection->chunkslice.len = 0;
projection->memslice.start = 0;
projection->memslice.stop = 0;
projection->memslice.stride = 1;
projection->memslice.len = 0;
}
/* Goal:
Create a vector of projections wrt a slice and a sequence of chunks.
*/
int
NCZ_compute_per_slice_projections(
struct Common* common,
int r, /* which dimension are we projecting? */
const NCZSlice* slice, /* the slice for which projections are computed */
const NCZChunkRange* range, /* range */
NCZSliceProjections* slp)
{
int stat = NC_NOERR;
size64_t index,slicecount;
size_t n;
/* Part fill the Slice Projections */
slp->r = r;
slp->range = *range;
slp->count = range->stop - range->start;
if((slp->projections = calloc(slp->count,sizeof(NCZProjection))) == NULL)
{stat = NC_ENOMEM; goto done;}
/* Compute the total number of output items defined by this slice
(equivalent to count as used by nc_get_vars) */
slicecount = ceildiv((slice->stop - slice->start), slice->stride);
if(slicecount < 0) slicecount = 0;
/* Iterate over each chunk that intersects slice to produce projection */
for(n=0,index=range->start;index<range->stop;index++,n++) {
if((stat = NCZ_compute_projections(common, r, index, slice, n, slp->projections)))
goto done; /* something went wrong */
}
done:
return stat;
}
/* Goal:create a vector of SliceProjection instances: one for each
slice in the top-level input. For each slice, compute a set
of projections from it wrt a dimension and a chunk size
associated with that dimension.
*/
int
NCZ_compute_all_slice_projections(
struct Common* common,
const NCZSlice* slices, /* the complete set of slices |slices| == R*/
const NCZChunkRange* ranges,
NCZSliceProjections* results)
{
int stat = NC_NOERR;
size64_t r;
for(r=0;r<common->rank;r++) {
/* Compute each of the rank SliceProjections instances */
NCZSliceProjections* slp = &results[r];
if((stat=NCZ_compute_per_slice_projections(
common,
r,
&slices[r],
&ranges[r],
slp))) goto done;
}
done:
return stat;
}
/**************************************************/
/* Utilities */
/* return 0 if slice is malformed; 1 otherwise */
static int
verifyslice(const NCZSlice* slice)
{
if(slice->stop < slice->start) return 0;
if(slice->stride <= 0) return 0;
if((slice->stop - slice->start) > slice->len) return 0;
return 1;
}
void
NCZ_clearsliceprojections(int count, NCZSliceProjections* slpv)
{
if(slpv != NULL) {
int i;
for(i=0;i<count;i++) {
NCZSliceProjections* slp = &slpv[i];
nullfree(slp->projections);
}
}
}
#if 0
static void
clearallprojections(NCZAllProjections* nap)
{
if(nap != NULL) {
int i;
for(i=0;i<nap->rank;i++)
nclistfreeall(&nap->allprojections[i].projections);
}
}
#endif