using fsspec adaptive-readahead cache

[raj-prince]

Changes on top of - adaptive-readahead cache in fsspec. fsspec/filesystem_spec#2093

[gemini-code-assist]

Code Review

This pull request migrates the custom prefetching logic from gcsfs to fsspec by replacing the local prefetcher implementation with compatibility re-exports from fsspec.prefetch, and updates the default cache type to adaptive_readahead. The reviewer identified three key issues: a potential AttributeError in ZonalFile due to the removal of _prefetch_engine, the need for a specific fallback cache type (readahead_chunked) for ZonalFile when adaptive_readahead is unavailable, and a potential ImportError when running with older versions of fsspec that lack the prefetch module.

[gemini-code-assist]

Since self._prefetch_engine has been completely removed from GCSFile.__init__, calling _fetch_range on a ZonalFile instance will raise an AttributeError because ZonalFile._fetch_range (in gcsfs/zonal_file.py line 193) still accesses self._prefetch_engine directly. Defining self._prefetch_engine = None here maintains backward compatibility and prevents this runtime error.

Suggested change

	self.consistency = consistency
	self.checker = get_consistency_checker(consistency)
	self.consistency = consistency
	self.checker = get_consistency_checker(consistency)
	self._prefetch_engine = None

[gemini-code-assist]

When adaptive_readahead is not available in the environment, falling back to "readahead" is appropriate for GCSFile, but for ZonalFile it should fall back to "readahead_chunked" (which was its previous default and is optimized for zonal/gRPC paths). Falling back to "readahead" for ZonalFile could lead to performance degradation.

Suggested change

	if "r" in mode and cache_type == "adaptive_readahead":
	if cache_type not in fsspec.core.caches:
	warnings.warn(
	"fsspec adaptive_readahead cache is unavailable in this environment; "
	"falling back to readahead"
	)
	cache_type = "readahead"
	if "r" in mode and cache_type == "adaptive_readahead":
	if cache_type not in fsspec.core.caches:
	fallback = "readahead_chunked" if type(self).__name__ == "ZonalFile" else "readahead"
	warnings.warn(
	f"fsspec adaptive_readahead cache is unavailable in this environment; "
	f"falling back to {fallback}"
	)
	cache_type = fallback

[gemini-code-assist]

If gcsfs is used with an older version of fsspec that does not have fsspec.prefetch, importing BackgroundPrefetcher directly will raise an ImportError and crash concurrent downloads. Wrapping this import in a try-except block and falling back to sequential download (_get_file_request) provides graceful backward compatibility.

Suggested change

	from fsspec.prefetch import BackgroundPrefetcher
	try:
	from fsspec.prefetch import BackgroundPrefetcher
	except ImportError:
	warnings.warn(
	"fsspec.prefetch is unavailable in this environment; "
	"falling back to sequential download"
	)
	return await self._get_file_request(
	rpath, lpath, headers=headers, callback=callback, **kwargs
	)