It is inefficient for object stores to have meaningful path names (unlike posix). A more varied and random pattern is better.
We should use this as an opportunity to put all the search/understanding in a separate location.
Caringo has an NFS interface to the as well. We haven't made use of that yet but if in future there's a requirement to access the same objects you are writing now via NFS, then numbers of objects per "directory" ( which are designated by fake "/" in the object URL) and normal POSIX file naming rules apply.
So include some slashes in there to reduce the number of nodes is wise.
Need to think about a sensible strategy for grouping at sensible levels.
agstephens
commented
4 years ago
It is inefficient for object stores to have meaningful path names (unlike posix). A more varied and random pattern is better.
We should use this as an opportunity to put all the search/understanding in a separate location.
Caringo has an NFS interface to the as well. We haven't made use of that yet but if in future there's a requirement to access the same objects you are writing now via NFS, then numbers of objects per "directory" ( which are designated by fake "/" in the object URL) and normal POSIX file naming rules apply.
So include some slashes in there to reduce the number of nodes is wise.
Need to think about a sensible strategy for grouping at sensible levels.