import streamlit as st import time import pandas as pd import plotly.express as px import snowflake.connector import base64 from datetime import timedelta, datetime from cryptography.hazmat.primitives import serialization from cryptography.hazmat.backends import default_backend import concurrent.futures # Import SQL query functions. from house_ad_queries import ( get_main_query, get_flex_query, get_bidder_query, get_deal_query, get_ad_unit_query, get_browser_query, get_device_query, get_random_integer_query, get_hb_pb_query, get_hb_size_query, ) # Import the house ad section config. from house_ad_section_utils import update_section_generic # Import the NEXT_STEPS_INSTRUCTIONS at the top. from house_ad_instructions import NEXT_STEPS_INSTRUCTIONS # Initialize session state keys at the top so they only get set once. st.session_state.setdefault("query_run", False) st.session_state.setdefault("findings_messages", []) st.session_state.setdefault("key_findings_output", None) st.session_state.setdefault("query_df", None) st.session_state.setdefault("agg_df", None) st.session_state.setdefault("top_level_spike_time", None) # --- Helper Functions --- # def load_private_key(key_str): # """Load a PEM-formatted private key.""" # return serialization.load_pem_private_key( # key_str.encode("utf-8"), # password=None, # backend=default_backend() # ) # Use caching to avoid re-running the same query on every interaction. @st.cache_data(show_spinner=False) def cached_run_query( query, private_key_b64: str, user: str, account_identifier: str, warehouse: str, database: str, schema: str, role: str, ): # 1) Decode the base64‐encoded DER key der = base64.b64decode(private_key_b64) """Connect to Snowflake and execute the given query. Cached to reduce re-runs.""" # private_key_obj = load_private_key(key_str=private_key_str) conn = snowflake.connector.connect( user=user, account=account_identifier, warehouse=warehouse, database=database, schema=schema, role=role, private_key=der, ) cs = conn.cursor() cs.execute("ALTER SESSION SET STATEMENT_TIMEOUT_IN_SECONDS = 1800") cs.execute(query) results = cs.fetchall() columns = [col[0] for col in cs.description] df = pd.DataFrame(results, columns=columns) cs.close() conn.close() return df # --- Main Function for House Ad Spike Analysis --- def run_house_ad_spike_query( table, start_datetime, end_datetime, message_filter, campaign_id, private_key_str, user, account_identifier, warehouse, database, schema, role, client, ): """ Run the house ad spike query along with additional dimensions, generate key findings via OpenAI, and display the results. """ # --- Generate SQL Queries --- main_sql = get_main_query( table, start_datetime, end_datetime, message_filter, campaign_id ) flex_sql = get_flex_query( table, start_datetime, end_datetime, message_filter, campaign_id ) bidder_sql = get_bidder_query( table, start_datetime, end_datetime, message_filter, campaign_id ) deal_sql = get_deal_query( table, start_datetime, end_datetime, message_filter, campaign_id ) ad_unit_sql = get_ad_unit_query( table, start_datetime, end_datetime, message_filter, campaign_id ) browser_sql = get_browser_query( table, start_datetime, end_datetime, message_filter, campaign_id ) device_sql = get_device_query( table, start_datetime, end_datetime, message_filter, campaign_id ) random_integer_sql = get_random_integer_query( table, start_datetime, end_datetime, message_filter, campaign_id ) hb_pb_sql = get_hb_pb_query( table, start_datetime, end_datetime, message_filter, campaign_id ) hb_size_sql = get_hb_size_query( table, start_datetime, end_datetime, message_filter, campaign_id ) # --- Main Query Execution --- # Run query only if it hasn't been run already. if not st.session_state["query_run"]: try: start_main = time.time() with st.spinner("Connecting to Snowflake and running top-level query..."): df = cached_run_query( main_sql, private_key_str, user, account_identifier, warehouse, database, schema, role, ) elapsed_main = time.time() - start_main elapsed_minutes = int(elapsed_main // 60) elapsed_seconds = elapsed_main % 60 st.info( f"Top-level SQL query executed in {elapsed_minutes} minute(s) and {elapsed_seconds:.2f} seconds." ) # Process the results. df.columns = [col.upper() for col in df.columns] df.sort_values(by=["EST_HOUR", "EST_MINUTE"], inplace=True) df["timestamp"] = pd.to_datetime( df["EST_DATE"].astype(str) + " " + df["EST_HOUR"].astype(str).str.zfill(2) + ":" + df["EST_MINUTE"].astype(str).str.zfill(2) ) df["5min"] = df["timestamp"].dt.floor("5T") agg_df = df.groupby("5min", as_index=False)["CNT"].sum() st.session_state["query_df"] = df st.session_state["agg_df"] = agg_df st.session_state["query_run"] = True except Exception as e: st.error(f"Error during main query execution: {e}") return else: # Use stored data. df = st.session_state.get("query_df") agg_df = st.session_state.get("agg_df") # --- Display Main Query Results --- st.header("Top-Level Data") top_level_baseline = 30 agg_df["is_spike"] = agg_df.apply( lambda row: row["CNT"] > top_level_baseline, axis=1 ) spike_start = None consecutive = 0 for idx, row in agg_df.sort_values("5min").iterrows(): if row["is_spike"]: consecutive += 1 if consecutive == 2: spike_start = row["5min"] - timedelta(minutes=5) break else: consecutive = 0 if spike_start: msg = f"Top-Level: House ad increase detected starting around {spike_start.strftime('%I:%M %p')}." st.success(msg) else: msg = "Top-Level: No large, consistent spike detected in the current data." st.info(msg) # Append the message only once. findings_messages = st.session_state.setdefault("findings_messages", []) if msg not in findings_messages: findings_messages.append(msg) st.session_state["top_level_spike_time"] = spike_start with st.expander("Show Raw Data"): st.dataframe(df) with st.expander("Show Raw 5-Minute Aggregated Data with Spike Alert"): st.dataframe(agg_df) title_text = "House Ads Count by 5-Minute Interval" fig = px.line( agg_df, x="5min", y="CNT", title=title_text, labels={"5min": "Time", "CNT": "House Ads Count"}, ) fig.update_xaxes(tickformat="%I:%M %p") st.plotly_chart(fig, use_container_width=True) st.markdown("