Repository navigation
Conversation
added 2 commits
September 18, 2026 11:55
gnutls_record_send() and gnutls_record_recv() may return GNUTLS_E_AGAIN or GNUTLS_E_INTERRUPTED. Both are non-fatal, gnutls_error_is_fatal() returns 0 for them, and the call is meant to be repeated. csync2 passed the negative return straight up instead: conn_read() turned it into a short read and csync_recv_file() called csync_fatal(), while csync_send_file() saw rc != chunk and did the same. A large transfer therefore died on a condition GnuTLS reports as harmless, in practice after roughly 7.4 GiB over a single connection. Pushing 31.66 GB of mixed files to two peers took seven csync2 runs to converge; with this it takes one. Both paths need it. Patching only the sending side changed nothing, because the receive path then aborted on the identical condition. The send loop only advances the offset after a successful send, so a retry repeats the identical (pointer, length) pair, as the GnuTLS manual requires. The socket is blocking, no O_NONBLOCK is ever set on conn_fd_in/conn_fd_out, so the retry cannot spin. The non-TLS branch of WRITE() already loops over partial writes and retries EINTR; this makes the TLS branch behave the same.
csync2 moved file payloads in 512-byte chunks, which over TLS works out to roughly one record per 512 bytes of file data. 16384 is the maximum TLS record payload, so anything larger is split back down by the TLS layer and gains nothing: 64 KiB measured 947 s against 946 s for 16 KiB. On three nodes pushing 31.66 GB of 56525 mixed files to two peers, with the GNUTLS_E_AGAIN retry fix applied in both runs so neither aborts, the sync goes from 1938 s to 979 s. Block size is the only variable between the two. This depends on the retry fix. Without it any chunk above the TLS record limit fails on the first file, because gnutls_record_send() returns a short count that csync_send_file() treats as fatal. The chunk size in csync_rs_patch()'s local copy fallback is left at 512 bytes, it never touches the network. The two buffers in csync_rs_check() come to 32 KiB of stack together.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #46. Depends on #45, please merge that one first.
512 byte chunks mean roughly one TLS record per 512 bytes of file data. Raising
them to 16384, the maximum TLS record payload, took the same 31.66 GB sync from
1938 s to 979 s with the retry fix applied in both runs. 64 KiB gains nothing
over 16 KiB, since TLS splits anything larger back down.
Without the retry fix this fails on the first file, because
gnutls_record_send()returns a short count above the record limit andcsync_send_file()treats that as fatal.