TW
Interesting. So, which of `--files-cache=ctime,size,inode` does this trigger?
Hmm, maybe winbind does **not** produce cache misses (at least not if you always have it active), but it is just **winbind** being **very slow** compared to a normal `stat()`...
TODO: check if `borg create --numeric-ids ...` does uid/gid -> name lookups.
The normal `stat` processing at backup time does **not** do any uid/gid to user/group name lookups if `--numeric-owner` (correct option name - sorry, i gave an incorrect one in my...
So, guess we need an `strace` (with and without winbind nss) here to determine what's slow.
@detoxhby thanks for the infos, I'll do some related tests (but I don't have an AD member to test on). One idea for you might be to run strace and...
The lru cache works, as expected (tried with borg 1.1-maint branch, which is ~ 1.1.15 and python 3.7.3 - but the lru cache stuff did not change recently): ``` (borg-env)...
To get this output, I added a `print()` to some functions in `borg/helpers.py`, like e.g. this: ``` @lru_cache(maxsize=None) def uid2user(uid, default=None): try: user = pwd.getpwuid(uid).pw_name print(f"getpwuid({uid}) -> {user} [default:{default}]") return...
@biji good idea! Guess the libacl code will internally do id -> name lookups if we do not explicitly request ACL_TEXT_NUMERIC_IDS (and as it seems from this ticket, it does...
Until there is a better solution with improved ACL code, I will offer a workaround via #3955. So, if backup/restore of ACLs is not really required, one will be able...