Hangs with Master only running

Question

Have been 4.7.25 with replication capabilities. Starting Master only running and no slaves hangs after a while.
The db put() hangs.
On attaching debugger the thread is waiting on mutex lock.
Tried to see who has locked the mutex and another thread is locked the mutes and the trace is
#0 0x00007ff7415c0f4b in read () from /lib/libpthread.so.0
#1 0x00007ff6bd0db7dc in __repmgr_select_loop (env=0x7ff6b84611f0) at ../dist/../repmgr/repmgr_posix.c:666
#2 0x00007ff6bd0dbea5 in __repmgr_select_thread (args=0x7ff6b84611f0) at ../dist/../repmgr/repmgr_sel.c:56
#3 0x00007ff7415ba3ea in start_thread () from /lib/libpthread.so.0
#4 0x00007ff741123cbd in clone () from /lib/libc.so.6
#5 0x0000000000000000 in ?? ()
(gdb) up
#1 0x00007ff6bd0db7dc in __repmgr_select_loop (env=0x7ff6b84611f0) at ../dist/../repmgr/repmgr_posix.c:666
666                    if (read(db_rep->read_pipe, buf, sizeof(buf)) <= 0) {
Looks like when the queue is full we try to wake up this thread from __repmgr_wake_main_thread().
This function is sending one byte only in pipe. and the read is trying to read 10 bytes as buf is defined as buf[0].
Is this the cause of the his hang up ??
Can somebody at Oracle check this out.
Thanks

Answer 1

Thanks for taking time to look into this.
We are running on linux Redhat and Ubuntu servers.
What we noticed after adding fprintf statements to function that is writing to pipe and the function reading from pipe, we noticed that in the beginning that the write and read works properly.
But after while the select() call triggers a read ready on the reap_pipe handle (spurious one). This leads to read hanging and the log also indicates that no write has happened earlier for the read to succeed.
I patched it up in a different way and this SEEMS to solve this issue. I would like you to verify if this has any ramifications.
I am using the db-4.7.25.NC .tar.gz and these line numbers below correspond to this download image from your site.
1. File : repmgr/repmgr_posix.c : original line number 327
Set the read pipe to non blocking mode
db_rep->read_pipe = file_desc[0];
db_rep->write_pipe = file_desc[1];
/* PATCH: to not block on read upon spurious select() */
__repmgr_set_nonblocking(db_rep->read_pipe);
2. Same File : repmgr/repmgr_posix.c : original line number 767
Since we made it non-blocking we need handle this condition appropriately.
if (FD_ISSET((u_int)db_rep->read_pipe, &reads)) {
* PATCH : this handle has been made non-blocking now
* to avoid getting blocked on read due to spurious select()
* wake-ups; this was causing the mutex to be locked
* and no db operations was possible after that
if (read(db_rep->read_pipe, buf, sizeof(buf)) <= 0) {
switch (ret = errno) {
case EINTR:
case EWOULDBLOCK:
if (db_rep->finished) {
ret = 0;
goto out;
} else {
break; /* simply retry later*/
default:
ret = errno;
goto out;
} else if (db_rep->finished) {
ret = 0;
goto out;
The above two changes made this problem go away and we do not see the master hanging any more.
Could you please verify if this is logical and does not cause any issue with the way the system is supposed to work.
Thanks
Raj