fix: Addresses ZeroDivisionError when materializing file source with …

…same timestamps (#2551) * Add docs for Go feature server Signed-off-by: Felix Wang <wangfelix98@gmail.com> * Update go feature server docs Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Address review components Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Fix Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Revert Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Fix Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Update comment Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Fix Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Fix Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * Revert indent Signed-off-by: Kevin Zhang <kzhang@tecton.ai> * fix comment Signed-off-by: Kevin Zhang <kzhang@tecton.ai> Co-authored-by: Felix Wang <wangfelix98@gmail.com>
feast-dev · Apr 20, 2022 · 5539c51 · 5539c51
1 parent 39d3f9d
commit 5539c51
Show file tree

Hide file tree

Showing 2 changed files with 19 additions and 5 deletions.
diff --git a/docs/reference/feature-servers/go-feature-retrieval.md b/docs/reference/feature-servers/go-feature-retrieval.md
@@ -10,7 +10,7 @@ The Go Feature Retrieval component currently only supports Redis and Sqlite as o
 
 ## Installation
 
-As long as you are running macOS or linux x86 with python version 3.7-3.10, the go component comes pre-compiled when you run install feast.
+As long as you are running macOS or linux, on x86, with python version 3.7-3.10, the go component comes pre-compiled when you install feast.
 
 For developers, if you want to build from source, run `make compile-go-lib` to build and compile the go server.
 

diff --git a/sdk/python/feast/infra/offline_stores/file.py b/sdk/python/feast/infra/offline_stores/file.py
@@ -299,11 +299,25 @@ def evaluate_offline_job():
                 if created_timestamp_column
                 else [event_timestamp_column]
             )
+            # try-catch block is added to deal with this issue https://github.com/dask/dask/issues/8939.
+            # TODO(kevjumba): remove try catch when fix is merged upstream in Dask.
+            try:
+                if created_timestamp_column:
+                    source_df = source_df.sort_values(by=created_timestamp_column,)
+
+                source_df = source_df.sort_values(by=event_timestamp_column)
+
+            except ZeroDivisionError:
+                # Use 1 partition to get around case where everything in timestamp column is the same so the partition algorithm doesn't
+                # try to divide by zero.
+                if created_timestamp_column:
+                    source_df = source_df.sort_values(
+                        by=created_timestamp_column, npartitions=1
+                    )
 
-            if created_timestamp_column:
-                source_df = source_df.sort_values(by=created_timestamp_column)
-
-            source_df = source_df.sort_values(by=event_timestamp_column)
+                source_df = source_df.sort_values(
+                    by=event_timestamp_column, npartitions=1
+                )
 
             source_df = source_df[
                 (source_df[event_timestamp_column] >= start_date)