[issue37355] SSLSocket.read does a GIL round-trip for every 16KB TLS record
Change by Josh Snyder : -- keywords: +patch pull_requests: +24203 stage: -> patch review pull_request: https://github.com/python/cpython/pull/25478 ___ Python tracker <https://bugs.python.org/issue37355> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue37355] SSLSocket.read does a GIL round-trip for every 16KB TLS record
New submission from Josh Snyder : Background: SSLSocket.read drops the GIL and performs exactly one successful call to OpenSSL's `SSL_read`, whose documentation states "At most the contents of one record will be returned". TLS records are at most 16KB, so high throughput (especially multithreaded) TLS reception can become bottlenecked on the GIL. Proposal: For non-blocking sockets, call SSL_read in a loop until the user-supplied limit is reached or no bytes are available on the socket. I don't know of a way to safely improve performance for blocking sockets. Initial testing: I performed initial testing using 32 threads pinned to 16 cores, downloading and re-assembling a single 140270MB file from a "real world" TLS sender. This resulted in a 4x increase in throughput, a 6.6x reduction in voluntary context switches, a 3.5x reduction in system time. User time did increase by 43%, so the overall reduction in CPU usage was only 2.67x. before after wall clock time (s):29.637 7.116 user time (s) : 8.79312.584 system time (s): 105.11830.010 user + system time (s) : 113.91142.594 cpu utilization (%): 384 599 voluntary switches : 1,653,065 248,484 speed (MB/s) : 4733 19712 My git branch (currently a draft) is at https://github.com/hashbrowncipher/cpython/commits/faster_tls -- assignee: christian.heimes components: SSL messages: 346156 nosy: christian.heimes, josnyder priority: normal severity: normal status: open title: SSLSocket.read does a GIL round-trip for every 16KB TLS record type: performance ___ Python tracker <https://bugs.python.org/issue37355> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue35046] logging.StreamHandler performs two syscalls when one would do
Change by Josh Snyder : -- keywords: +patch pull_requests: +9381 stage: -> patch review ___ Python tracker <https://bugs.python.org/issue35046> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue35046] logging.StreamHandler performs two syscalls when one would do
New submission from Josh Snyder : logging.StreamHandler contains the following code: stream.write(msg) stream.write(self.terminator) stream.flush() When sys.stderr (or whatever other stream) is unbuffered, this results in two system calls and allows log records from different processes to concatenate on the same line in the output stream (followed by multiple newlines). This issue is new in Python 3.7, as stdout and stderr became "truly unbuffered" (cf. #30404). As a simple solution, I believe the following would fix the issue and also be backward compatible: stream.write(msg + self.terminator) stream.flush() -- messages: 328269 nosy: josnyder priority: normal severity: normal status: open title: logging.StreamHandler performs two syscalls when one would do type: behavior versions: Python 3.7 ___ Python tracker <https://bugs.python.org/issue35046> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue32475] Add ability to query number of buffered bytes available on buffered I/O
Josh Snyder added the comment: I've opened PR #7947 with another approach to this issue. In my use-case, an HTTP client makes a request and uses buffered I/O to parse the response headers. I would like to hand off the response socket to an extension module for use with raw I/O syscalls. The BufferedReader likely already contains some of the response body, which I would like to access before handing the socket off. -- nosy: +josnyder ___ Python tracker <https://bugs.python.org/issue32475> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
[issue32475] Add ability to query number of buffered bytes available on buffered I/O
Change by Josh Snyder : -- pull_requests: +7556 stage: -> patch review ___ Python tracker <https://bugs.python.org/issue32475> ___ ___ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
