Add chunk specification to the katdal export

HISTORY.rst

@@ -4,6 +4,7 @@ History
 
 X.Y.Z (YYYY-MM-DD)
 ------------------
+* Add chunk specification to ``dask-ms katdal import`` (:pr:`318`)
 * Add a ``dask-ms katdal import`` application for exporting SARAO archive data directly to zarr (:pr:`315`)
 * Define dask-ms command line applications with click (:pr:`317`)
 * Make poetry dev and docs groups optional (:pr:`316`)

daskms/apps/convert.py

@@ -1,4 +1,3 @@
-import ast
 from argparse import ArgumentTypeError
 from collections import defaultdict
 import logging
@@ -8,37 +7,11 @@
 
 from daskms.apps.formats import TableFormat, CasaFormat
 from daskms.fsspec_store import DaskMSStore
+from daskms.utils import parse_chunks_dict
 
 log = logging.getLogger(__name__)
 
 
-class ChunkTransformer(ast.NodeTransformer):
-    def visit_Module(self, node):
-        if len(node.body) != 1 or not isinstance(node.body[0], ast.Expr):
-            raise ValueError("Module must contain a single expression")
-
-        expr = node.body[0]
-
-        if not isinstance(expr.value, ast.Dict):
-            raise ValueError("Expression must contain a dictionary")
-
-        return self.visit(expr).value
-
-    def visit_Dict(self, node):
-        keys = [self.visit(k) for k in node.keys]
-        values = [self.visit(v) for v in node.values]
-        return {k: v for k, v in zip(keys, values)}
-
-    def visit_Name(self, node):
-        return node.id
-
-    def visit_Tuple(self, node):
-        return tuple(self.visit(v) for v in node.elts)
-
-    def visit_Constant(self, node):
-        return node.n
-
-
 NONUNIFORM_SUBTABLES = ["SPECTRAL_WINDOW", "POLARIZATION", "FEED", "SOURCE"]
 
 
@@ -88,7 +61,7 @@ def _check_exclude_columns(ctx, param, value):
 
 
 def parse_chunks(ctx, param, value):
-    return ChunkTransformer().visit(ast.parse(value))
+    return parse_chunks_dict(value)
 
 
 def col_converter(ctx, param, value):

daskms/apps/katdal_import.py

@@ -1,5 +1,7 @@
 import click
 
+from daskms.utils import parse_chunks_dict
+
 
 @click.group()
 @click.pass_context
@@ -58,10 +60,16 @@ def convert(self, value, param, ctx):
     "'K,B,G'. Use 'default' for L1 + L2 and 'all' for "
     "all available products.",
 )
-def _import(ctx, rdb_url, no_auto, pols_to_use, applycal, output_store):
+@click.option(
+    "--chunks",
+    callback=lambda c, p, v: parse_chunks_dict(v),
+    default="{time: 10}",
+    help="Chunking values to apply to each dimension",
+)
+def _import(ctx, rdb_url, output_store, no_auto, pols_to_use, applycal, chunks):
     """Export an observation in the SARAO archive to zarr formation
 
     RDB_URL is the SARAO archive link"""
     from daskms.experimental.katdal import katdal_import
 
-    katdal_import(rdb_url, output_store, no_auto, applycal)
+    katdal_import(rdb_url, output_store, no_auto, applycal, chunks)

daskms/experimental/katdal/katdal_import.py

@@ -1,10 +1,14 @@
+import logging
 import os
 import urllib
 
 import dask
 
+from daskms.fsspec_store import DaskMSStore
 from daskms.utils import requires
 
+log = logging.getLogger(__file__)
+
 try:
     import katdal
     from katdal.dataset import DataSet
@@ -31,20 +35,25 @@ def default_output_name(url):
 
 
 @requires("pip install dask-ms[katdal]", import_error)
-def katdal_import(url: str, out_store: str, no_auto: bool, applycal: str):
+def katdal_import(url: str, out_store: str, no_auto: bool, applycal: str, chunks: dict):
     if isinstance(url, str):
         dataset = katdal.open(url, appycal=applycal)
     elif isinstance(url, DataSet):
         dataset = url
     else:
         raise TypeError(f"{url} must be a string or a katdal DataSet")
 
-    facade = XarrayMSV2Facade(dataset, no_auto=no_auto)
+    facade = XarrayMSV2Facade(dataset, no_auto=no_auto, chunks=chunks)
     main_xds, subtable_xds = facade.xarray_datasets()
 
     if not out_store:
         out_store = default_output_name(url)
 
+    out_store = DaskMSStore(out_store)
+    if out_store.exists():
+        log.warn("Removing previously existing %s", out_store)
+        out_store.rm("", recursive=True)
+
     writes = [
         xds_to_zarr(main_xds, out_store),
         *(xds_to_zarr(ds, f"{out_store}::{k}") for k, ds in subtable_xds.items()),

daskms/experimental/katdal/msv2_facade.py

@@ -58,10 +58,22 @@ def to_mjds(timestamp: Timestamp):
     return timestamp.to_mjd() * 24 * 60 * 60
 
 
+DEFAULT_TIME_CHUNKS = 100
+DEFAULT_CHAN_CHUNKS = 4096
+DEFAULT_CHUNKS = {"time": DEFAULT_TIME_CHUNKS, "chan": DEFAULT_CHAN_CHUNKS}
+
+
 class XarrayMSV2Facade:
     """Provides a simplified xarray Dataset view over a katdal dataset"""
 
-    def __init__(self, dataset: DataSet, no_auto: bool = True, row_view: bool = True):
+    def __init__(
+        self,
+        dataset: DataSet,
+        no_auto: bool = True,
+        row_view: bool = True,
+        chunks: dict = None,
+    ):
+        self._chunks = chunks or DEFAULT_CHUNKS
         self._dataset = dataset
         self._no_auto = no_auto
         self._row_view = row_view
@@ -81,6 +93,10 @@ def _main_xarray_factory(self, field_id, state_id, scan_index, scan_state, targe
         time_utc = dataset.timestamps
         t_chunks, chan_chunks, cp_chunks = dataset.vis.dataset.chunks
 
+        # Override time and channel chunking
+        t_chunks = self._chunks.get("time", t_chunks)
+        chan_chunks = self._chunks.get("chan", chan_chunks)
+
         # Modified Julian Date in Seconds
         time_mjds = np.asarray([to_mjds(t) for t in map(Timestamp, time_utc)])
 
@@ -110,7 +126,14 @@ def _main_xarray_factory(self, field_id, state_id, scan_index, scan_state, targe
 
         flags = DaskLazyIndexer(dataset.flags, (), (rechunk, flag_transpose))
         weights = DaskLazyIndexer(dataset.weights, (), (rechunk, weight_transpose))
-        vis = DaskLazyIndexer(dataset.vis, (), transforms=(vis_transpose,))
+        vis = DaskLazyIndexer(
+            dataset.vis,
+            (),
+            transforms=(
+                rechunk,
+                vis_transpose,
+            ),
+        )
 
         time = da.from_array(time_mjds[:, None], chunks=(t_chunks, 1))
         ant1 = da.from_array(cp_info.ant1_index[None, :], chunks=(1, cpi.shape[0]))

daskms/tests/test_utils.py

@@ -7,6 +7,7 @@
 import pytest
 
 from daskms.utils import (
+    parse_chunks_dict,
     promote_columns,
     natural_order,
     table_path_split,
@@ -15,6 +16,18 @@
 )
 
 
+def test_parse_chunks_dict():
+    assert parse_chunks_dict("{row: 1000}") == {"row": 1000}
+    assert parse_chunks_dict("{row: 1000, chan: 64}") == {"row": 1000, "chan": 64}
+    assert parse_chunks_dict("{row: (10, 10), chan: (4, 4)}") == {
+        "row": (10, 10),
+        "chan": (4, 4),
+    }
+
+    with pytest.raises(SyntaxError):
+        parse_chunks_dict("row:1000}")
+
+
 def test_natural_order():
     data = [f"{i}.parquet" for i in reversed(range(20))]
     expected = [f"{i}.parquet" for i in range(20)]

daskms/utils.py

@@ -1,17 +1,14 @@
 # -*- coding: utf-8 -*-
 
+import ast
 from collections import OrderedDict
-import importlib.util
 import logging
 from pathlib import PurePath, Path
 import re
-import sys
 import time
 import inspect
 import warnings
 
-from dask.utils import funcname
-
 # The numpy module may disappear during interpreter shutdown
 # so explicitly import ndarray
 from numpy import ndarray
@@ -21,6 +18,37 @@
 log = logging.getLogger(__name__)
 
 
+class ChunkTransformer(ast.NodeTransformer):
+    def visit_Module(self, node):
+        if len(node.body) != 1 or not isinstance(node.body[0], ast.Expr):
+            raise ValueError("Module must contain a single expression")
+
+        expr = node.body[0]
+
+        if not isinstance(expr.value, ast.Dict):
+            raise ValueError("Expression must contain a dictionary")
+
+        return self.visit(expr).value
+
+    def visit_Dict(self, node):
+        keys = [self.visit(k) for k in node.keys]
+        values = [self.visit(v) for v in node.values]
+        return {k: v for k, v in zip(keys, values)}
+
+    def visit_Name(self, node):
+        return node.id
+
+    def visit_Tuple(self, node):
+        return tuple(self.visit(v) for v in node.elts)
+
+    def visit_Constant(self, node):
+        return node.n
+
+
+def parse_chunks_dict(chunks_str):
+    return ChunkTransformer().visit(ast.parse(chunks_str))
+
+
 def natural_order(key):
     return tuple(
         int(c) if c.isdigit() else c.lower() for c in re.split(r"(\d+)", str(key))