optimize_rasters.py

"""scripts/optimize_rasters.py

Convert some raster files to cloud-optimized GeoTiff for use with Terracotta.
"""

from typing import Any, Sequence, Iterator, Union
import os
import sys
import math
import warnings
import itertools
import contextlib
import tempfile
import logging
from pathlib import Path
import concurrent.futures

import click
import click_spinner
import rasterio
from rasterio.shutil import copy
from rasterio.io import DatasetReader, MemoryFile
from rasterio.errors import NotGeoreferencedWarning
from rasterio.vrt import WarpedVRT
from rasterio.enums import Resampling
from rasterio.env import GDALVersion
from rasterio.warp import calculate_default_transform

from terracotta.scripts.click_types import GlobbityGlob, PathlibPath

logger = logging.getLogger(__name__)

IN_MEMORY_THRESHOLD = 10980 * 10980

CACHEMAX = os.environ.get("GDAL_CACHEMAX", 1024 * 1024 * 512)  # 512 MB

GDAL_CONFIG = {
    "GDAL_TIFF_INTERNAL_MASK": True,
    "GDAL_TIFF_OVR_BLOCKSIZE": 256,
    "GDAL_CACHEMAX": CACHEMAX,
    "GDAL_SWATH_SIZE": 2 * CACHEMAX,
}

COG_PROFILE = {
    "count": 1,
    "driver": "GTiff",
    "interleave": "pixel",
    "tiled": True,
    "blockxsize": 256,
    "blockysize": 256,
    "photometric": "MINISBLACK",
    "ZLEVEL": 1,
    "ZSTD_LEVEL": 9,
    "BIGTIFF": "IF_SAFER",
}

RESAMPLING_METHODS = {
    "average": Resampling.average,
    "nearest": Resampling.nearest,
    "bilinear": Resampling.bilinear,
    "cubic": Resampling.cubic,
}


def _prefered_compression_method() -> str:
    if not GDALVersion.runtime().at_least("2.3"):
        return "DEFLATE"

    # check if we can use ZSTD (fails silently for GDAL < 2.3)
    dummy_profile = dict(driver="GTiff", height=1, width=1, count=1, dtype="uint8")
    try:
        with warnings.catch_warnings():
            warnings.simplefilter("ignore", NotGeoreferencedWarning)

            with MemoryFile() as memfile, memfile.open(
                compress="ZSTD", **dummy_profile
            ):
                pass

    except Exception as exc:
        if "missing codec" not in str(exc):
            raise
    else:
        return "ZSTD"

    return "DEFLATE"


def _get_vrt(src: DatasetReader, rs_method: int) -> WarpedVRT:
    from terracotta.drivers.geotiff_raster_store import GeoTiffRasterStore

    target_crs = GeoTiffRasterStore._TARGET_CRS
    vrt_transform, vrt_width, vrt_height = calculate_default_transform(
        src.crs, target_crs, src.width, src.height, *src.bounds
    )
    vrt = WarpedVRT(
        src,
        crs=target_crs,
        resampling=rs_method,
        transform=vrt_transform,
        width=vrt_width,
        height=vrt_height,
    )
    return vrt


@contextlib.contextmanager
def _named_tempfile(basedir: Union[str, Path]) -> Iterator[str]:
    fileobj = tempfile.NamedTemporaryFile(dir=str(basedir), suffix=".tif")
    fileobj.close()
    try:
        yield fileobj.name
    finally:
        os.remove(fileobj.name)


TemporaryRasterFile = _named_tempfile


def _output_file(output_folder: Path, input_file: Path) -> Path:
    return output_folder / input_file.with_suffix(".tif").name


def _optimize_single_raster(
    input_file: Path,
    output_folder: Path,
    reproject: bool,
    rs_method: Any,
    in_memory: Union[bool, None],
    compression: str,
    quiet: bool,
    progress_suffix: str,
) -> None:
    output_file = _output_file(output_folder, input_file)

    if not quiet:
        click.echo(f"\r{input_file.name} ... {progress_suffix}")

    with contextlib.ExitStack() as es, warnings.catch_warnings():
        warnings.filterwarnings("ignore", message="invalid value encountered.*")

        src = es.enter_context(rasterio.open(str(input_file)))

        if reproject:
            vrt = es.enter_context(_get_vrt(src, rs_method=rs_method))
        else:
            vrt = src

        profile = vrt.profile.copy()
        profile.update(COG_PROFILE)

        if in_memory is None:
            in_memory = vrt.width * vrt.height < IN_MEMORY_THRESHOLD

        if in_memory:
            memfile = es.enter_context(MemoryFile())
            dst = es.enter_context(memfile.open(**profile))
        else:
            tempraster = es.enter_context(TemporaryRasterFile(basedir=output_folder))
            dst = es.enter_context(rasterio.open(tempraster, "w", **profile))

        # iterate over blocks
        windows = list(dst.block_windows(1))

        for _, w in windows:
            block_data = vrt.read(window=w, indexes=[1])
            dst.write(block_data, window=w)
            block_mask = vrt.dataset_mask(window=w).astype("uint8")
            dst.write_mask(block_mask, window=w)

        # add overviews
        if not in_memory:
            # work around bug mapbox/rasterio#1497
            dst.close()
            dst = es.enter_context(rasterio.open(tempraster, "r+"))

        max_overview_level = math.ceil(
            math.log2(
                max(
                    dst.height // profile["blockysize"],
                    dst.width // profile["blockxsize"],
                    1,
                )
            )
        )

        if max_overview_level > 0:
            overviews = [2**j for j in range(1, max_overview_level + 1)]
            dst.build_overviews(overviews, rs_method)

            dst.update_tags(ns="rio_overview", resampling=rs_method.value)

        # copy to destination (this is necessary to push overviews to start of file)
        copy(
            dst,
            str(output_file),
            copy_src_overviews=True,
            compress=compression,
            **COG_PROFILE,
        )


@click.command(
    "optimize-rasters",
    short_help="Optimize a collection of raster files for use with Terracotta.",
)
@click.argument("raster-files", nargs=-1, type=GlobbityGlob(), required=True)
@click.option(
    "-o",
    "--output-folder",
    required=True,
    type=PathlibPath(file_okay=False, writable=True),
    help="Output folder for cloud-optimized rasters. Subdirectories will be flattened.",
)
@click.option(
    "--skip-existing", is_flag=True, default=False, help="Skip existing files"
)
@click.option(
    "--overwrite", is_flag=True, default=False, help="Force overwrite of existing files"
)
@click.option(
    "--resampling-method",
    type=click.Choice(list(RESAMPLING_METHODS.keys())),
    default="average",
    help="Resampling method for overviews",
    show_default=True,
)
@click.option(
    "--reproject",
    is_flag=True,
    default=False,
    show_default=True,
    help="Reproject raster file to Web Mercator for faster access",
)
@click.option(
    "--in-memory/--no-in-memory",
    default=None,
    help="Force processing raster in memory / not in memory [default: process in memory "
    f"if smaller than {IN_MEMORY_THRESHOLD // 1e6:.0f} million pixels]",
)
@click.option(
    "--compression",
    default="auto",
    type=click.Choice(["auto", "deflate", "lzw", "zstd", "none"]),
    help="Compression algorithm to use [default: auto (ZSTD if available, DEFLATE otherwise)]",
)
@click.option(
    "--nproc",
    default=1,
    type=click.INT,
    help="Number of processes to use for multi-core processing "
    "[default: 1, i.e., single-core processing] "
    "Set to -1 to use all available (logical) cores",
)
@click.option(
    "-q",
    "--quiet",
    is_flag=True,
    default=False,
    show_default=True,
    help="Suppress all output to stdout",
)
def optimize_rasters(
    raster_files: Sequence[Sequence[Path]],
    output_folder: Path,
    overwrite: bool = False,
    skip_existing: bool = False,
    resampling_method: str = "average",
    reproject: bool = False,
    in_memory: Union[bool, None] = None,
    compression: str = "auto",
    nproc: int = 1,
    quiet: bool = False,
) -> None:
    """Optimize a collection of raster files for use with Terracotta.

    First argument is a list of input files or glob patterns.

    Example:

        $ terracotta optimize-rasters rasters/*.tif -o cloud-optimized/

    Note that all rasters may only contain a single band.
    """
    if overwrite and skip_existing:
        raise click.BadOptionUsage(
            "--overwrite and --skip-existing",
            "Both --overwrite and --skip-existing flags are provided. "
            "These are mutually exclusive. Please provide at max one of them",
        )

    raster_files_flat = set(itertools.chain.from_iterable(raster_files))

    if not raster_files_flat:
        click.echo("No files given")
        return

    rs_method = RESAMPLING_METHODS[resampling_method]

    if compression == "auto":
        compression = _prefered_compression_method()

    total_pixels = 0
    raster_files_to_skip = set()
    for f in raster_files_flat:
        if not f.is_file():
            raise click.BadParameter(f"Input raster {f!s} is not a file")

        output_file = _output_file(output_folder, f)
        if output_file.is_file():
            if not (skip_existing or overwrite):
                raise click.BadParameter(
                    f"Output file {f!s} exists (use --overwrite or --skip-existing)"
                )
            elif skip_existing:
                raster_files_to_skip.add(f)

        with rasterio.open(str(f), "r") as src:
            if src.count > 1 and not quiet:
                click.echo(
                    f"Warning: raster file {f!s} has more than one band. "
                    "Only the first one will be used.",
                    err=True,
                )
            total_pixels += src.height * src.width
    raster_files_to_optimize = sorted(raster_files_flat - raster_files_to_skip)

    output_folder.mkdir(exist_ok=True)

    if nproc == -1:
        nproc = os.cpu_count() or 1  # Default to 1 if `cpu_count` returns None

    if not quiet:
        files_str = "file" if len(raster_files_to_optimize) == 1 else "files"
        processes_str = "process" if nproc == 1 else "processes"
        click.echo(
            f"Optimizing {len(raster_files_to_optimize)} {files_str} on {nproc} {processes_str}"
        )

    with contextlib.ExitStack() as outer_env:
        outer_env.enter_context(
            click_spinner.spinner(
                beep=False, disable=quiet, force=False, stream=sys.stdout
            )
        )
        outer_env.enter_context(rasterio.Env(**GDAL_CONFIG))

        if nproc > 1:
            executor = outer_env.enter_context(
                concurrent.futures.ProcessPoolExecutor(max_workers=nproc)
            )

            futures = {
                executor.submit(
                    _optimize_single_raster,
                    input_file,
                    output_folder,
                    reproject,
                    rs_method,
                    in_memory,
                    compression,
                    quiet,
                    f"({i}/{len(raster_files_to_optimize)})",
                ): input_file
                for i, input_file in enumerate(raster_files_to_optimize, start=1)
            }

            for future in concurrent.futures.as_completed(futures):
                try:
                    future.result()
                except Exception as exc:
                    raise RuntimeError(
                        f"Error while optimizing file {futures[future]}"
                    ) from exc
        else:  # Single-core; run in the current process
            for i, input_file in enumerate(raster_files_to_optimize, start=1):
                _optimize_single_raster(
                    input_file,
                    output_folder,
                    reproject,
                    rs_method,
                    in_memory,
                    compression,
                    quiet,
                    f"({i}/{len(raster_files_to_optimize)})",
                )