From e89c654b3aa8671ddd36da0c8404071abe276ec4 Mon Sep 17 00:00:00 2001
From: Aaron Fenyes <aaron.fenyes@fareycircles.ooo>
Date: Tue, 24 Sep 2024 16:17:12 -0700
Subject: [PATCH] Allow stale values requests to be repeated

This addresses issue #154, using the same technique as PR #127.
---
 flaskr/nscope/models.py |  8 +++----
 flaskr/nscope/views.py  | 53 ++++++++++++++++++++++++++++++++++-------
 2 files changed, 48 insertions(+), 13 deletions(-)

diff --git a/flaskr/nscope/models.py b/flaskr/nscope/models.py
index fee9205..3ae5384 100644
--- a/flaskr/nscope/models.py
+++ b/flaskr/nscope/models.py
@@ -37,13 +37,13 @@ class Sequence(db.Model):
     # Postgres reserved word, so we use a different name.
     shift = db.Column(db.Integer, unique=False, nullable=False, default=0)
     values = db.Column(db.ARRAY(db.String), unique=False, nullable=True)
-    values_requested = db.Column(db.Boolean, nullable=False, default=False)
     raw_refs = db.Column(db.String, unique=False, nullable=True)
     backrefs = db.Column(db.ARRAY(db.String), unique=False, nullable=True)
     ref_count = db.Column(db.Integer, nullable=True, default=None)
-    # The start time of the last attempt to fetch metadata, in nanoseconds since
-    # the UNIX epoch. A PostgreSQL BigInteger is eight bytes, including sign, so
-    # this should work until the early 2260s
+    # The start times of the last attempt to fetch values and metadata, in
+    # nanoseconds since the UNIX epoch. A PostgreSQL BigInteger is eight bytes,
+    # including sign, so this should work until the early 2260s
+    values_req_time = db.Column(db.BigInteger, nullable=True, default=None)
     meta_req_time = db.Column(db.BigInteger, nullable=True, default=None)
     # Sadly, multidimensional arrays can't vary in dimension
     # so we store factorization arrays as strings
diff --git a/flaskr/nscope/views.py b/flaskr/nscope/views.py
index 0bfcbd3..572462f 100644
--- a/flaskr/nscope/views.py
+++ b/flaskr/nscope/views.py
@@ -235,16 +235,38 @@ def fetch_values(oeis_id):
         OEIS (if it has not already been), and returns a Sequence object
         with the values filled in.
     """
-    # First check if it is in the database
     seq = find_oeis_sequence(oeis_id)
-    # See if we already have the values:
-    if seq.values is not None: return seq
-    # See if getting them is in progress:
-    if seq.values_requested:
-        return LookupError("Value fetching for {oeis_id} in progress.")
-    seq.values_requested = True
+    if seq.values is not None:
+        # We already have the values in the database, so we just return them
+        return seq
+    
+    our_req_time = time.time_ns()
+    last_req_time = seq.values_req_time
+    if last_req_time is not None:
+        # we chose `max_wait = 3` pretty haphazardly. it matches the minimum
+        # `max_wait` for a metadata request that has some refs downloaded
+        # already, and it's longer than the time it took to download the A000521
+        # b-file on Glen's connection (two seconds for 4 MB)
+        waited = (our_req_time - last_req_time) / 1e9
+        max_wait = 3
+        if waited < max_wait:
+            return LookupError(
+                f"Values for {oeis_id} were already requested {waited:.1f} "
+                "seconds ago. A new request can be made if the old one takes "
+                f"longer than {max_wait:.1f} seconds."
+            )
+    
+    #---------------------------------------------------------------------------
+    # if we've gotten this far, we don't have the values in the database yet,
+    # and we don't think any other thread is likely to come back with them
+    #---------------------------------------------------------------------------
+
+    # Record the time we set out to fetch the values, so later threads can
+    # judge how likely we are to ever come back
+    seq.values_req_time = our_req_time
     db.session.commit()
-    # Now try to get it from the OEIS:
+    
+    # Try to get the b-file from the OEIS:
     b_text = oeis_get(f'/{oeis_id}/b{oeis_id[1:]}.txt', json=False)
     # Test for 404 error. Hat tip StackOverflow user Lukasa
     #   https://stackoverflow.com/a/19343099
@@ -283,7 +305,20 @@ def fetch_values(oeis_id):
     if not seq.name:
         seq.name = name or placeholder_name(oeis_id)
     seq.shift = first
-    db.session.commit()
+    
+    # We write what we've found to the database in the following situations:
+    #
+    # - No more recent thread has set out to fetch the same metadata
+    #
+    # - A more recent thread has set out to fetch the same metadata, but we got
+    #   back before any other thread did
+    #
+    # This is equivalent to the condition in the `if` statement below because
+    # the only way for `seq.values_req_time == our_req_time` to be false is for
+    # a more recent thread to have overwritten the request time
+    if seq.values_req_time == our_req_time or seq.values is None:
+        db.session.commit()
+    
     return seq
 
 def fetch_factors(oeis_id, num_elements = -1, timeout = 10):