Hi, I noticed that the TUNA Anaconda mirror does not seem to provide CEP-16 sharded repodata for conda-forge, while the upstream conda-forge channel already provides it.
For example, upstream conda-forge has:
curl -I https://conda.anaconda.org/conda-forge/win-64/repodata_shards.msgpack.zst
which returns:
HTTP/1.1 200 OK
Content-Type: binary/octet-stream
Content-Length: 414167
and:
curl -I https://conda.anaconda.org/conda-forge/noarch/repodata_shards.msgpack.zst
also returns:
HTTP/1.1 200 OK
Content-Type: binary/octet-stream
Content-Length: 929729
However, the TUNA mirror currently returns 404 for the same file:
curl -I https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/win-64/repodata_shards.msgpack.zst
returns:
while the traditional flat repodata file is available:
curl -I https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/win-64/repodata.json
returns:
This causes recent versions of mamba / libmamba to print warnings such as:
Shard Index for conda-forge/noarch not available, falling back to flat repodata
Shard Index for conda-forge/win-64 not available, falling back to flat repodata
The installation still works, but the client falls back to downloading/parsing the much larger flat repodata.json, which can be noticeably slower for large subdirs such as conda-forge/win-64.
From the mirror sync script, it seems that the current logic only downloads traditional metadata files:
repodata.json
repodata.json.bz2
repodata.json.zst
current_repodata.json
but does not sync:
repodata_shards.msgpack.zst
or the referenced shard files under info.shards_base_url, commonly ./shards/.
Would it be possible to add support for syncing CEP-16 sharded repodata for conda cloud channels, especially conda-forge?
Thanks.
Hi, I noticed that the TUNA Anaconda mirror does not seem to provide CEP-16 sharded repodata for
conda-forge, while the upstreamconda-forgechannel already provides it.For example, upstream
conda-forgehas:which returns:
and:
also returns:
However, the TUNA mirror currently returns 404 for the same file:
returns:
while the traditional flat repodata file is available:
returns:
This causes recent versions of
mamba/libmambato print warnings such as:The installation still works, but the client falls back to downloading/parsing the much larger flat
repodata.json, which can be noticeably slower for large subdirs such asconda-forge/win-64.From the mirror sync script, it seems that the current logic only downloads traditional metadata files:
but does not sync:
or the referenced shard files under
info.shards_base_url, commonly./shards/.Would it be possible to add support for syncing CEP-16 sharded repodata for conda cloud channels, especially
conda-forge?Thanks.