Problem updating the rendering when overwriting a file

[odhondt]

Titiler is a great tool but I was wondering: is there a way to clear the cache so that when a file is overwritten the preview is updated?

I am using the default app in jupyter notebooks and even if I re-execute the cell, the image stays the same after a file is overwritten.
So far the only workaround I found is to add a small random value to the bounds of the rescale parameter (or to write the file with a different name every time, which is not really practical in my use case.)

[vincentsarago]

Hi @odhondt

This has been previously discussed here but I can't find the ticket.

you can check rasterio/rasterio#2181 and rasterio/rasterio#2209

[odhondt]

Hi @vincentsarago
Thanks.
So I understand it has something to do with GDAL VSI cache.
What I still don't get is how to fix that using TiTiler.

[vincentsarago]

There is no direct fix in TiTiler itself.

You could may do something like

from fastapi import Query
from rasterio._path import _parse_path as parse_path
from typing_extensions import Annotated


def EnvParams(url: Annotated[str, Query(description="Dataset URL")]) -> str:
    return {"CPL_VSIL_CURL_NON_CACHED": parse_path(url).as_vsi()}


from titiler.core.factory import TilerFactory
from fastapi import FastAPI

app = FastAPI()
cog = TilerFactory(environment_dependency=EnvParams)
app.include_router(cog.router)

ref: https://github.com/developmentseed/tilebench/blob/main/tilebench/viz.py#L140

You may also add VSI_CACHE=False and GDAL_CACHEMAX=0 in your environment to force GDAL not too cache anything

☝️ this will impact the performance.

note: You may do something smarter by first check if the file has changed (e.g by checking the time of the file creation) before setting CPL_VSIL_CURL_NON_CACHED env.

[odhondt]

That's super interesting. So that means serving a custom app instead of using the default one, right?

[vincentsarago]

yes

[odhondt]

thanks