net_ib: return ncclSuccess if read roceTypePath failed and errno is EINVAL. #890

lijunli · lijunli · commit dc445aa5dacf · 2024-09-05T11:08:42.000+08:00
diff --git a/src/transport/net_ib.cc b/src/transport/net_ib.cc
@@ -248,6 +248,10 @@ static ncclResult_t ncclIbRoceGetVersionNum(const char* deviceName, int portNum,
   close(fd);
 
   if (ret == -1) {
+    WARN("NET/IB: read failed: %s, roceTypePath: %s", strerror(errno), roceTypePath);
+    if (errno == EINVAL) {
+      return ncclSuccess;
+    }
     return ncclSystemError;
   }
 

Original file line number	Diff line number	Diff line change
`@@ -248,6 +248,10 @@ static ncclResult_t ncclIbRoceGetVersionNum(const char* deviceName, int portNum,`
`248`	`248`	`close(fd);`
`249`	`249`
`250`	`250`	`if (ret == -1) {`
	`251`	`+ WARN("NET/IB: read failed: %s, roceTypePath: %s", strerror(errno), roceTypePath);`
	`252`	`+ if (errno == EINVAL) {`
	`253`	`+ return ncclSuccess;`
	`254`	`+ }`
`251`	`255`	`return ncclSystemError;`
`252`	`256`	`}`
`253`	`257`