Win32 asynchronous queries hang on connection error

[ged]

Original report by Rafał Bigaj (Bitbucket: rafalbigaj, GitHub: rafalbigaj).

---

I encountered the following problems on win32 system with all version of pg (0.9.0, 0.9.1, 0.10.0, 0.10.1 and 0.11.0) on Ruby 1.9:

During asynchronous queries the CPU usage is 100% - rb_w32_thread_select does not wait any time on foreign socket (not created with ruby socket API). The only way I found to workaround the problem is simple waiting 1ms before each select if the PG is busy.

PQConsumeInput result is not interpreted in pgconn_block what makes the waiting loop infinite (PQisBusy returns always TRUE) if connection error occurs

pgconn_async_exec can block the interpreter for a long time on initial pgconn_get_last_result (pgconn_block is not called before)

I have committed the fixes for above problems to my fork at:
https://bitbucket.org/rafalbigaj/ruby-pg/changeset/17a6a01c1725

[ged]

Original comment by Michael Granger (Bitbucket: ged, GitHub: ged).

---

Thanks for the patch. I'll look it over this weekend.

[ged]

Original comment by Michael Granger (Bitbucket: ged, GitHub: ged).

---

Applying the patch as-is breaks the build for me:

../../../../ext/pg.c:2602:39: error: operator '||' has no right operand
../../../../ext/pg.c: In function ‘pgconn_block’:
../../../../ext/pg.c:2608: error: ‘zerotime’ undeclared (first use in this function)
../../../../ext/pg.c:2608: error: (Each undeclared identifier is reported only once
../../../../ext/pg.c:2608: error: for each function it appears in.)

It's likely just a matter of a missing defined() at line 2602.

I'll fix that and get this applied. Do you have a minimal test case that demonstrates the problem? I can probably figure it out myself, but I'm wondering if it's possible to write a test for it, and thought if you already had something you used, I'd have a headstart.

[ged]

Original comment by Anonymous.

---

After running into performance problems I found out that the previous patch that added a small wait (1ms) after each PQisBusy is not a good idea. I have committed the next patch:

https://bitbucket.org/rafalbigaj/ruby-pg/changeset/fc4109b2ddba

which uses Winsock events and rb_w32_wait_events.

The proposed solution is faster about 20-30% on WIN32 than the original code.

[ged]

Original comment by Rafał Bigaj (Bitbucket: rafalbigaj, GitHub: rafalbigaj).

---

I have attached the simple script that I uses to test if the connection problem still exist.

During the script is running I release my network connection. On the original code I frequently encounter hangs.

How to write automated test for it?

[ged]

Original comment by Rafał Bigaj (Bitbucket: rafalbigaj, GitHub: rafalbigaj).

---

One commit more:
https://bitbucket.org/rafalbigaj/ruby-pg/changeset/2bc78e46c1b6
that fixes memory leaks (forgot to close opened event).

[ged]

Original comment by Michael Granger (Bitbucket: ged, GitHub: ged).

---

Fixes #66 Win32 asynchronous queries hang on connection error