Don't spuriously report FD_SETSIZE exhaustion on Windows.

Starting on 2023-08-03, this intermittently terminated a "pgbench -C"
test in CI.  It could affect a high-client-count "pgbench" without "-C".
While parallel reindexdb and vacuumdb reach the same problematic check,
sufficient client count and/or connection turnover is less plausible for
them.  Given the lack of examples from the buildfarm or from manual
builds, reproducing this must entail rare operating system
configurations.  Also correct the associated error message, which was
wrong for non-Windows.  Back-patch to v12, where the pgbench check first
appeared.  While v11 vacuumdb has the problematic check, reaching it
with typical vacuumdb usage is implausible.

Reviewed by Thomas Munro.

Discussion: https://postgr.es/m/CA+hUKG+JwvTNdcyJTriy9BbtzF1veSRQ=9M_ZKFn9_LqE7Kp7Q@mail.gmail.com

diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 63487f349f593e7042aae1ca3ac18ee13910f502..f6e4d04ea1ab04756c6abfc1aaa2ad38b00c36b1 100644 (file)
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7136,15 +7136,22 @@ clear_socket_set(socket_set *sa)
 static void
 add_socket_to_set(socket_set *sa, int fd, int idx)
 {
+   /* See connect_slot() for background on this code. */
+#ifdef WIN32
+   if (sa->fds.fd_count + 1 >= FD_SETSIZE)
+   {
+       pg_log_fatal("too many concurrent database clients for this platform: %d",
+                    sa->fds.fd_count + 1);
+       exit(1);
+   }
+#else
    if (fd < 0 || fd >= FD_SETSIZE)
    {
-       /*
-        * Doing a hard exit here is a bit grotty, but it doesn't seem worth
-        * complicating the API to make it less grotty.
-        */
-       pg_log_fatal("too many client connections for select()");
+       pg_log_fatal("socket file descriptor out of range for select(): %d",
+                    fd);
        exit(1);
    }
+#endif
    FD_SET(fd, &sa->fds);
    if (fd > sa->maxfd)
        sa->maxfd = fd;

diff --git a/src/fe_utils/parallel_slot.c b/src/fe_utils/parallel_slot.c
index dcdad9e30c5cc562ef2d11dc76d9cf09c9c78eed..8d558064343c17d039ef5fdb674e3bdd8751f6b9 100644 (file)
--- a/src/fe_utils/parallel_slot.c
+++ b/src/fe_utils/parallel_slot.c
@@ -297,11 +297,40 @@ connect_slot(ParallelSlotArray *sa, int slotno, const char *dbname)
    slot->connection = connectDatabase(sa->cparams, sa->progname, sa->echo, false, true);
    sa->cparams->override_dbname = old_override;
 
-   if (PQsocket(slot->connection) >= FD_SETSIZE)
+   /*
+    * POSIX defines FD_SETSIZE as the highest file descriptor acceptable to
+    * FD_SET() and allied macros.  Windows defines it as a ceiling on the
+    * count of file descriptors in the set, not a ceiling on the value of
+    * each file descriptor; see
+    * https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-select
+    * and
+    * https://learn.microsoft.com/en-us/windows/win32/api/winsock/ns-winsock-fd_set.
+    * We can't ignore that, because Windows starts file descriptors at a
+    * higher value, delays reuse, and skips values.  With less than ten
+    * concurrent file descriptors, opened and closed rapidly, one can reach
+    * file descriptor 1024.
+    *
+    * Doing a hard exit here is a bit grotty, but it doesn't seem worth
+    * complicating the API to make it less grotty.
+    */
+#ifdef WIN32
+   if (slotno >= FD_SETSIZE)
    {
-       pg_log_fatal("too many jobs for this platform");
+       pg_log_fatal("too many jobs for this platform: %d", slotno);
        exit(1);
    }
+#else
+   {
+       int         fd = PQsocket(slot->connection);
+
+       if (fd >= FD_SETSIZE)
+       {
+           pg_log_fatal("socket file descriptor out of range for select(): %d",
+                        fd);
+           exit(1);
+       }
+   }
+#endif
 
    /* Setup the connection using the supplied command, if any. */
    if (sa->initcmd)