| Summary: | SSL_Asynch_Stream_Test is failing in all builds | ||
|---|---|---|---|
| Product: | ACE | Reporter: | Johnny Willemsen <jwillemsen> |
| Component: | SSL Wrappers | Assignee: | DOC Center Support List (internal) <tao-support> |
| Status: | NEW --- | ||
| Severity: | normal | ||
| Priority: | P3 | ||
| Version: | 5.6.1 | ||
| Hardware: | All | ||
| OS: | All | ||
| Bug Depends on: | 1426, 1959, 1960 | ||
| Bug Blocks: | |||
|
Description
Johnny Willemsen
2007-12-04 08:16:28 CST
The client connect is hanging inside ssl_connect(). I've made a number of fixes to the test itself, and have now run into, I believe, basic problems in the proactor code itself. 1. There's no notify done when new operations are initiated and this leaves some (most?) platforms with aio_suspend() calls that will never notice completions because the operations were initiated after aio_suspend() was called. This is the case currently on HP-UX SSL_Asynch_Stream_Test. 2. The aiocb list is modified both for new ops and for completion handling without any synchronization. This causes random weirdness which varies to some extent by platform. These issues have been reported in various forms over the years; I've referenced items 1959, 1960, 1426 as other manifestations of these issues. When addressing these, make sure all problems noted are resolved. An additional note from info gathered debugging this on Windows, which doesn't suffer from the POSIX proactor deficiencies noted previously. There are problems happening when the connections reach the point they should shut down. There's no SSL shutdown happening, apparently because there's a write queued when the shutdown is asked for. Not sure why. |