Retry loop with backoff in test_server_start

Signed-off-by: Cole Miller <[email protected]>
canonical · Feb 22, 2024 · ba76637 · ba76637
1 parent d408763
commit ba76637
Show file tree

Hide file tree

Showing 5 changed files with 22 additions and 6 deletions.
diff --git a/include/dqlite.h b/include/dqlite.h
@@ -217,7 +217,8 @@ DQLITE_API DQLITE_EXPERIMENTAL void dqlite_server_destroy(
 enum {
 	DQLITE_ERROR = 1, /* Generic error */
 	DQLITE_MISUSE,    /* Library used incorrectly */
-	DQLITE_NOMEM      /* A malloc() failed */
+	DQLITE_NOMEM,     /* A malloc() failed */
+	DQLITE_AGAIN      /* Retryable error */
 };
 
 /**

diff --git a/src/raft.h b/src/raft.h
@@ -55,7 +55,8 @@ enum {
 	RAFT_INVALID,      /* Invalid parameter */
 	RAFT_UNAUTHORIZED, /* No access to a resource */
 	RAFT_NOSPACE,      /* Not enough space on disk */
-	RAFT_TOOMANY       /* Some system or raft limit was hit */
+	RAFT_TOOMANY,      /* Some system or raft limit was hit */
+	RAFT_AGAIN         /* Retryable error */
 };
 
 /**

diff --git a/src/raft/uv_fs.c b/src/raft/uv_fs.c
@@ -774,8 +774,11 @@ static int probeAsyncIO(int fd, size_t size, bool *ok, char *errmsg)
 	rv = UvOsIoSetup(1, &ctx);
 	if (rv != 0) {
 		UvOsErrMsg(errmsg, "io_setup", rv);
-		/* UNTESTED: in practice this should fail only with ENOMEM */
-		return RAFT_IOERR;
+		if (rv == -EAGAIN) {
+			return RAFT_AGAIN;
+		} else {
+			return RAFT_IOERR;
+		}
 	}
 
 	/* Allocate the write buffer */

diff --git a/src/server.c b/src/server.c
@@ -110,7 +110,7 @@ int dqlite__init(struct dqlite_node *d,
 	if (rv != 0) {
 		snprintf(d->errmsg, DQLITE_ERRMSG_BUF_SIZE, "raft_init(): %s",
 			 raft_errmsg(&d->raft));
-		return DQLITE_ERROR;
+		return (rv == RAFT_AGAIN) ? DQLITE_AGAIN : DQLITE_ERROR;
 	}
 	/* TODO: expose these values through some API */
 	raft_set_election_timeout(&d->raft, 3000);

diff --git a/test/lib/server.c b/test/lib/server.c
@@ -61,7 +61,18 @@ void test_server_start(struct test_server *s, const MunitParameter params[])
 {
 	int rv;
 
-	rv = dqlite_node_create(s->id, s->address, s->dir, &s->dqlite);
+	struct timespec ts = {.tv_sec = 0, .tv_nsec = 1000000};
+	do {
+		rv = dqlite_node_create(s->id, s->address, s->dir, &s->dqlite);
+		if (rv != 0) {
+			dqlite_node_destroy(s->dqlite);
+		}
+		if (rv != DQLITE_AGAIN) {
+			break;
+		}
+		nanosleep(&ts, NULL);
+		ts.tv_nsec *= 2;
+	} while (ts.tv_nsec < 1000000000);
 	munit_assert_int(rv, ==, 0);
 
 	rv = dqlite_node_set_bind_address(s->dqlite, s->address);