Tvheadend

There's 2 seconds delay to detect, if EPG signalize the running status. It seems that the running status is delivered a bit late in your case, thus the recording started.

You may try modify line 'if (start_time + 2 >= gclk()) {' and increase the value there.

It may be also worth to check, if EPG has those big running state signalling window (--trace tbl-eit):

diff --git a/src/epggrab/module/eit.c b/src/epggrab/module/eit.c
index 4304f3b76..c0e8a43e6 100644
--- a/src/epggrab/module/eit.c
+++ b/src/epggrab/module/eit.c
@@ -743,6 +743,7 @@ running:
       case 4:  run = EPG_RUNNING_NOW;   break;
       default: run = EPG_RUNNING_STOP;  break;
       }
+      tvhtrace(mod->subsys, "running for %p set to %d", ebc, run);
       *save |= epg_broadcast_set_running(ebc, run);
     } else if (sect == 1 && running != 2 && running != 3 && running != 4) {
     }

My suggestion is a bit wrong, but the key is why 'real_start <= gclk()' is true, so the epg_running is set. The real_start should be time without the warmup subsctraction.

This patch fixes my issue, however I don't know whether it will break #5256.

--- dvr_rec.c.orig      2018-12-13 17:15:41.212915850 +0000
+++ dvr_rec.c   2018-12-13 17:22:56.630831452 +0000
@@ -1587,13 +1587,13 @@
   tvhtrace(LS_DVR, "%s - recoding thread started for \"%s\"",
            idnode_uuid_as_str(&de->de_id, ubuf), lang_str_get(de->de_title, NULL));
   if (!running_disabled && de->de_bcast) {
-    real_start = gclk();
+    time_t now = gclk();
     switch (de->de_bcast->running) {
     case EPG_RUNNING_PAUSE:
-      atomic_set_time_t(&de->de_running_pause, real_start);
+      atomic_set_time_t(&de->de_running_pause, now);
       /* fall through */
     case EPG_RUNNING_NOW:
-      atomic_set_time_t(&de->de_running_start, real_start);
+      atomic_set_time_t(&de->de_running_start, now);
       break;
     }
   }

Thanks, it seems like variable misuse.