diff --git a/fink_mm/conf/fink_mm.conf b/fink_mm/conf/fink_mm.conf
index 3455f2e1..a0a7f86e 100644
--- a/fink_mm/conf/fink_mm.conf
+++ b/fink_mm/conf/fink_mm.conf
@@ -13,7 +13,7 @@ hdfs_gcn_storage=/user/roman.le-montagner/gcn_storage/raw
 # They can be in local FS (/path/ or files:///path/) or
 # in distributed FS (e.g. hdfs:///path/).
 # Be careful though to have enough disk space!
-online_ztf_data_prefix=fink_mm/test/test_data/ztf_test
+online_ztf_data_prefix=fink_mm/test/test_data/ztf_test/online/science
 
 # Prefix path on disk to save GRB join ZTF data (work for both online and offline).
 online_grb_data_prefix=fink_mm/test/test_output
diff --git a/fink_mm/ztf_join_gcn.py b/fink_mm/ztf_join_gcn.py
index 45f68031..bd0f4b66 100644
--- a/fink_mm/ztf_join_gcn.py
+++ b/fink_mm/ztf_join_gcn.py
@@ -185,30 +185,19 @@ def load_dataframe(
         # connection to the ztf science stream
         ztf_alert = connect_to_raw_database(
             ztf_path
-            + "/online/science/year={}/month={}/day={}".format(
-                night[0:4], night[4:6], night[6:8]
-            ),
+            + "/year={}/month={}/day={}".format(night[0:4], night[4:6], night[6:8]),
             ztf_path
-            + "/online/science/year={}/month={}/day={}".format(
-                night[0:4], night[4:6], night[6:8]
-            ),
+            + "/year={}/month={}/day={}".format(night[0:4], night[4:6], night[6:8]),
             latestfirst=False,
         )
 
-        userschema = spark.read.option("mergeSchema", True).parquet(gcn_path).schema
-        gcn_alert = (
-            spark.readStream.format("parquet")
-            .schema(userschema)
-            .option("basePath", gcn_path)
-            .option(
-                "path",
-                gcn_path + "/year={}/month={}/day=*?*".format(night[0:4], night[4:6]),
-            )
-            .option("latestFirst", True)
-            .option("mergeSchema", True)
-            .load()
+        # load all GCN since the beginning
+        gcn_alert = connect_to_raw_database(
+            gcn_path,
+            gcn_path,
+            latestfirst=False,
         )
-        # keep gcn emitted during the day time until the end of the stream (17:00 Paris Time)
+        # keep gcn emitted between the last day time and the end of the current stream (17:00 Paris Time)
         cur_time = Time(f"{night[0:4]}-{night[4:6]}-{night[6:8]}")
         last_time = cur_time - timedelta(hours=7)  # 17:00 Paris time yesterday
         end_time = cur_time + timedelta(hours=17)  # 17:00 Paris time today
@@ -414,11 +403,11 @@ def gcn_pre_join(
 
 def ztf_join_gcn_stream(
     mm_mode: DataMode,
-    ztf_datapath_prefix: str,
+    ztf_dataframe: DataFrame,
+    gcn_dataframe: DataFrame,
     gcn_datapath_prefix: str,
     night: str,
     NSIDE: int,
-    time_window: int,
     hdfs_adress: str,
     ast_dist: float,
     pansstar_dist: float,
@@ -433,10 +422,10 @@ def ztf_join_gcn_stream(
     ----------
     mm_mode : DataMode
         run this function in streaming or offline mode.
-    ztf_datapath_prefix : string
-        the prefix path where are stored the ztf alerts.
-    gcn_datapath_prefix : string
-        the prefix path where are stored the gcn alerts.
+    ztf_dataframe : DataFrame
+        streaming dataframe containing the ztf alerts.
+    gcn_dataframe : DataFrame
+        streaming dataframe containing the gcn alerts.
     night : string
         the processing night
     NSIDE: String
@@ -472,14 +461,6 @@ def ztf_join_gcn_stream(
         "science2mm_{}_{}{}{}".format(job_name, night[0:4], night[4:6], night[6:8])
     )
 
-    ztf_dataframe, gcn_dataframe = load_dataframe(
-        spark,
-        ztf_datapath_prefix,
-        gcn_datapath_prefix,
-        night,
-        int(time_window),
-        mm_mode,
-    )
     ztf_dataframe = ztf_pre_join(
         ztf_dataframe, ast_dist, pansstar_dist, pansstar_star_score, gaia_dist, NSIDE
     )
@@ -564,6 +545,8 @@ def ztf_join_gcn(
         the maximum active time in second of the streaming process
     tinterval : int
         the processing interval time in second between the data batch
+    time_window : int
+        number of day in the past to load the gcn
     hdfs_adress: string
         HDFS adress used to instanciate the hdfs client from the hdfs package
     ast_dist: float
@@ -625,13 +608,22 @@ def ztf_join_gcn(
     """
     logger = init_logging()
 
+    ztf_dataframe, gcn_dataframe = load_dataframe(
+        spark,
+        ztf_datapath_prefix,
+        gcn_datapath_prefix,
+        night,
+        int(time_window),
+        mm_mode,
+    )
+
     df_join_mm, spark = ztf_join_gcn_stream(
         mm_mode,
-        ztf_datapath_prefix,
+        ztf_dataframe,
+        gcn_dataframe,
         gcn_datapath_prefix,
         night,
         NSIDE,
-        time_window,
         hdfs_adress,
         ast_dist,
         pansstar_dist,