is there a memory leak?

[vincentsarago]

Dear users, I'm trying to debug some possible memory leak issue with rio-tiler (this subject has been brought many time over the rasterio repo and it seems that is there is a leak it might be at users application level, e.g rio-tiler). Before raising a new issue in rasterio I would like to be sure that's it's not in rio-tiler 😅

Where ?

• rio_tiler.reader.read
• rio_tiler.models.ImageData
• rio_tiler.tasks.create_tasks
• rio_tiler.mosaic.*

I'm out of ideas/clues of what is going on so if someone wants to help, it will be really appreciated 🙏

python scripts: memory_mosaic.py

from rio_tiler.io import Reader
from rio_tiler.mosaic import mosaic_reader
from random import sample
import morecantile
import rasterio
from rio_tiler.models import ImageData
from memory_profiler import profile

from rio_tiler.mosaic.methods import LowestMethod

assets = [
    "tests/fixtures/mosaic_value_1.tif",
    "tests/fixtures/mosaic_value_2.tif",
    "tests/fixtures/mosaic_value_1.tif",
    "tests/fixtures/mosaic_value_2.tif",
]

tilematrixset = morecantile.tms.get("WebMercatorQuad")
w, s, e, n = [-76, 44.93, -71.33, 47.1]
minzoom = 7
maxzoom = 9

extrema = {}
for zoom in range(minzoom, maxzoom + 1):
    ul_tile = tilematrixset.tile(w, n, zoom)
    lr_tile = tilematrixset.tile(e, s, zoom)
    extrema[zoom] = {
        "x": {"min": ul_tile.x, "max": lr_tile.x + 1},
        "y": {"min": ul_tile.y, "max": lr_tile.y + 1},
    }

def _reader(path: str, x: int, y: int, z: int, **kwargs):
    with Reader(path) as src:
        return src.tile(x, y, z, **kwargs)


@profile()
def mosaic(x, y, z):
    # while True:
    im, _ = mosaic_reader(
        assets,
        _reader,
        x,
        y,
        z,
        threads=1,
        tilesize=4000,
        pixel_selection=LowestMethod,
    )
    im = None

    return True


@profile()
def image_from_list(x, y, z):
    imgs = []
    for asset in assets:
        try:
            im = _reader(asset, x, y, z, tilesize=4000)
            imgs.append(im)
        except:
            pass

    img = ImageData.create_from_list(imgs)
    imgs = None
    img = None

    return True


@profile()
def images(x, y, z):
    for asset in assets:
        try:
            im = _reader(asset, x, y, z, tilesize=4000)
            im = None
        except:
            pass

    return True

@profile()
def image(x, y, z):
    with Reader(assets[0]) as src:
        im = src.tile(x, y, z, tilesize=4000)
        im = None

    return True



if __name__ == '__main__':
    z = sample(range(minzoom, maxzoom + 1), 1)[0]
    x = sample(range(extrema[z]["x"]["min"], extrema[z]["x"]["max"]), 1)[0]
    y = sample(range(extrema[z]["y"]["min"], extrema[z]["y"]["max"]), 1)[0]

    _ = image(x, y, z)
    _ = images(x, y, z)
    _ = mosaic(x, y, z)
    _ = image_from_list(x, y, z)

results

python -m memory_profiler memory_mosaic.py
Filename: memory_mosaic.py

Line #    Mem usage    Increment  Occurrences   Line Contents
=============================================================
    83    191.2 MiB    191.2 MiB           1   @profile()
    84                                         def image(x, y, z):
    85    198.2 MiB      7.0 MiB           1       with Reader(assets[0]) as src:
    86    533.8 MiB    335.6 MiB           1           im = src.tile(x, y, z, tilesize=4000)
    87    533.8 MiB      0.0 MiB           1           im = None
    88                                         
    89    533.8 MiB      0.0 MiB           1       return True


Filename: memory_mosaic.py

Line #    Mem usage    Increment  Occurrences   Line Contents
=============================================================
    72    533.8 MiB    533.8 MiB           1   @profile()
    73                                         def images(x, y, z):
    74    581.2 MiB     -2.1 MiB           5       for asset in assets:
    75    581.2 MiB      0.0 MiB           4           try:
    76    581.2 MiB     45.2 MiB           4               im = _reader(asset, x, y, z, tilesize=4000)
    77    581.2 MiB     -2.1 MiB           4               im = None
    78                                                 except:
    79                                                     pass
    80                                         
    81    579.1 MiB     -2.1 MiB           1       return True


Filename: memory_mosaic.py

Line #    Mem usage    Increment  Occurrences   Line Contents
=============================================================
    37    579.1 MiB    579.1 MiB           1   @profile()
    38                                         def mosaic(x, y, z):
    39                                             # while True:
    40    856.8 MiB    277.8 MiB           2       im, _ = mosaic_reader(
    41    579.1 MiB      0.0 MiB           1           assets,
    42    579.1 MiB      0.0 MiB           1           _reader,
    43    579.1 MiB      0.0 MiB           1           x,
    44    579.1 MiB      0.0 MiB           1           y,
    45    579.1 MiB      0.0 MiB           1           z,
    46    579.1 MiB      0.0 MiB           1           threads=1,
    47    579.1 MiB      0.0 MiB           1           tilesize=4000,
    48    579.1 MiB      0.0 MiB           1           pixel_selection=LowestMethod,
    49                                             )
    50    856.8 MiB      0.0 MiB           1       im = None
    51                                         
    52    856.8 MiB      0.0 MiB           1       return True


Filename: memory_mosaic.py

Line #    Mem usage    Increment  Occurrences   Line Contents
=============================================================
    55    857.0 MiB    857.0 MiB           1   @profile()
    56                                         def image_from_list(x, y, z):
    57    857.0 MiB      0.0 MiB           1       imgs = []
    58   1042.9 MiB      0.0 MiB           5       for asset in assets:
    59    912.4 MiB      0.0 MiB           4           try:
    60   1042.9 MiB    185.9 MiB           4               im = _reader(asset, x, y, z, tilesize=4000)
    61   1042.9 MiB      0.0 MiB           4               imgs.append(im)
    62                                                 except:
    63                                                     pass
    64                                         
    65   1498.0 MiB    455.2 MiB           1       img = ImageData.create_from_list(imgs)
    66   1498.0 MiB      0.0 MiB           1       imgs = None
    67    948.7 MiB   -549.3 MiB           1       img = None
    68                                         
    69    948.7 MiB      0.0 MiB           1       return True

The issue was reported by a titiler-pgstac application user, resulting in memory increase like 👇

[wildintellect]

We decided to see if Valgrind could provide something more useful. @chuckwondo ran Scalene and it was showing that the memory usage is almost entirely outside python itself.

valgrind --tool=memcheck --leak-check=full --suppressions=valgrind-python.supp --log-file=minimal.valgrind.log python memory_mosaic.py

minimal.valgrind.log

[vincentsarago]

👀 rasterio/rasterio#2932