Hello,
IIUC, it is likely a bug of GnuPG.
When a server with higher load, it takes time for a process of gpg-agent
to start listen(2)ing the connection than the value of
SECS_TO_WAIT_FOR_AGENT. It's five seconds, hard-coded.
Last year, I pushed a change to remove possible race condition around
the initial connect to gpg-agent.
I don't know if this patch fixes the particular problem of sbuild, but,
it should improve the situation, hopefully, much.
--
commit b1c56cf9e2bb51abfd47747128bd2a6285ed1623
Author: NIIBE Yutaka
Date: Wed Jul 24 15:15:32 2019 +0900
common: Use gnupg_spawn_process_fd to invoke gpg-agent/dirmngr.
* common/asshelp.c (start_new_gpg_agent): Call gnupg_spawn_process_fd
and gnupg_wait_process.
(start_new_dirmngr): Likewise.
--
With --daemon option, gpg-agent/dirmngr detaches by itself.
Signed-off-by: NIIBE Yutaka
diff --git a/common/asshelp.c b/common/asshelp.c
index 5209ea6cf..600774330 100644
--- a/common/asshelp.c
+++ b/common/asshelp.c
@@ -492,8 +492,13 @@ start_new_gpg_agent (assuan_context_t *r_ctx,
if (!(err = lock_spawning (, gnupg_homedir (), "agent", verbose))
&& assuan_socket_connect (ctx, sockname, 0, 0))
{
- err = gnupg_spawn_process_detached (program? program : agent_program,
- argv, NULL);
+ pid_t pid;
+
+ err = gnupg_spawn_process_fd (program? program : agent_program,
+argv, -1, -1, -1, );
+ if (!err)
+err = gnupg_wait_process (program? program : agent_program,
+ pid, 1, NULL);
if (err)
log_error ("failed to start agent '%s': %s\n",
agent_program, gpg_strerror (err));
@@ -627,7 +632,12 @@ start_new_dirmngr (assuan_context_t *r_ctx,
if (!(err = lock_spawning (, gnupg_homedir (), "dirmngr", verbose))
&& assuan_socket_connect (ctx, sockname, 0, 0))
{
- err = gnupg_spawn_process_detached (dirmngr_program, argv, NULL);
+ pid_t pid;
+
+ err = gnupg_spawn_process_fd (dirmngr_program, argv,
+-1, -1, -1, );
+ if (!err)
+err = gnupg_wait_process (dirmngr_program, pid, 1, NULL);
if (err)
log_error ("failed to start the dirmngr '%s': %s\n",
dirmngr_program, gpg_strerror (err));
