Goal: retrieve only the events that have changed (added, updated, or deleted) since the last run, without downloading the entire dataset each time.

Context: The AXSMarine API exposes two GraphQL endpoints:

polygonEvents – the active events.
deletedPolygonEvents – the events that have been removed.

1. Prerequisites

Step

Details

Resources

1️⃣	Access Token	Obtain your bearer token from AXSMarine and replace the placeholder `INSERT MY TOKEN` in the script.
2️⃣	Python 3.9+	The script uses `requests` and `pandas`.
3️⃣	Persistence	A local CSV cache is used; you can swap it for a database or cloud storage if desired.
4️⃣	Timestamp handling	`lastUpdated` is inclusive; subtract 24 h from the last modification time to avoid missing boundary events.

2. API Architecture

Plain text

 # 1. polygonEvents
type Query {
  polygonEvents(
    first: Int,
    last: Int,
    before: String,
    after: String,
    isOpen: Boolean,
    polygonIds: [Int],
    polygonTypes: [String],
    entryDate: RangeDate,
    outDate: RangeDate,
    entryDraft: RangeFloat,
    entryHeading: RangeInt,
    entrySpeed: RangeFloat,
    outDraft: RangeFloat,
    outHeading: RangeInt,
    outSpeed: RangeFloat,
    duration: RangeInt,
    lastUpdated: RangeDate,
    vesselBuilt: RangeDate,
    vesselBeam: RangeFloat,
    vesselLoa: RangeFloat,
    vesselDraft: RangeFloat,
    vesselDwt: RangeInt,
    vesselTeu: RangeInt,
    vesselCubic: RangeInt,
    vesselIds: [Int],
    vesselImos: [Int],
    vesselSegments: [String],
    vesselTypes: [String],
    vesselSubtypes: [String],
    vesselLinerServiceIds: [Int],
    vesselLinerRegionIds: [Int]
  ): PolygonEventCursorConnection

# 2. deletedPolygonEvents
type Query {
  deletedPolygonEvents(
    first: Int,
    last: Int,
    before: String,
    after: String,
    polygonIds: [Int],
    polygonTypes: [String],
    deletedAt: RangeDate,
    vesselIds: [Int],
    vesselImos: [Int],
    vesselSegments: [String],
    vesselTypes: [String],
    vesselSubtypes: [String]
  ): deletedPolygonEventCursorConnection
}

Pagination – Each response returns a pageInfo (endCursor, startCursor) and a list of edges. Use after to iterate over the entire set.

3. Workflow Overview

Load the local cache (or initialize an empty dataframe).
Determine the update window:
1. If a file exists, the last run is its modification time minus 24 h.
2. Otherwise, this is the first run, and we’ll request all events.
Query polygonEvents with filters polygonTypes, entryDate, lastUpdated.
Loop over pages until the number of records is < pageSize.
Merge the new data into the dataframe and deduplicate on _id.
Query deletedPolygonEvents for the same window (deletedAt).
Remove rows that _id appear in the deletion list.
Save the final DataFrame to a CSV file.

4. Complete Python Example

Tip: Store your token in a .env file or use environment variables; the script uses it in plain text for demonstration.

Python

 #!/usr/bin/env python3
import os
import json
import logging
import requests
import pandas as pd
from datetime import datetime, timedelta

# --------------------------------------------------------------------
# 1️⃣  Configuration
# ------------------------------------------------------------------
TOKEN = "INSERT MY TOKEN"          # ← Replace with your token
HEADERS = {"Authorization": f"Bearer {TOKEN}"}
BASE_URL = "https://apihub.axsmarine.com/global/events/v1"

# Example: only retrieve "shipyard" events.
POLYGON_TYPES = ["shipyard"]
ENTRY_START_DATE = "2025-01-01"     # ISO‑8601
CSV_FILE = "2025_shipyard_events.csv"

# ------------------------------------------------------------------
# 2️⃣  Load cache (if present)
# ------------------------------------------------------------------
last_execution_time: str | None = None
data = pd.DataFrame()

if os.path.exists(CSV_FILE):
    # Ensure we don't miss events updated the same day.
    last_exec_ts = datetime.fromtimestamp(os.path.getmtime(CSV_FILE)) - timedelta(days=1)
    last_execution_time = last_exec_ts.strftime("%Y-%m-%d")
    data = pd.read_csv(CSV_FILE)

# ------------------------------------------------------------------
# 3️⃣  Generic GraphQL query helper
# ------------------------------------------------------------------
def gql_query(query: str, variables: dict) -> dict:
    """Execute a GraphQL request and return the JSON payload."""
    logging.debug(f"Request variables: {variables}")
    resp = requests.post(BASE_URL, json={"query": query, "variables": variables}, headers=HEADERS)
    resp.raise_for_status()
    return resp.json()["data"]

# ------------------------------------------------------------------
# 4️⃣  Retrieve active polygon events
# ------------------------------------------------------------------
def fetch_polygon_events() -> pd.DataFrame:
    """Fetch active events, paginating until exhaustion."""
    page_size = 5000
    variables = {
        "pageSize": page_size,
        "polygonTypes": POLYGON_TYPES,
        "entryDate": {"from": ENTRY_START_DATE},
        "lastUpdated": {"from": last_execution_time} if last_execution_time else None,
    }

    query = """
    query polygonEvents(
        $pageSize: Int,
        $afterCursor: String,
        $polygonTypes: [String],
        $entryDate: RangeDate,
        $lastUpdated: RangeDate
    ) {
        polygonEvents(
            first: $pageSize,
            after: $afterCursor,
            polygonTypes: $polygonTypes,
            entryDate: $entryDate,
            lastUpdated: $lastUpdated
        ) {
            pageInfo { endCursor }
            edges {
                node {
                    _id
                    polygon { _id, name }
                    vessel { _id, imo, name, type }
                    entryAis { time }
                    outAis   { time }
                    lastUpdated
                }
            }
        }
    }
    """

    all_rows = []
    cursor = None
    while True:
        variables["afterCursor"] = cursor
        payload = gql_query(query, variables)
        edges = payload["polygonEvents"]["edges"]
        rows = pd.json_normalize([e["node"] for e in edges])
        all_rows.append(rows)

        if len(edges) < page_size:
            break
        cursor = payload["polygonEvents"]["pageInfo"]["endCursor"]

    if all_rows:
        df = pd.concat(all_rows, ignore_index=True)
        return df
    return pd.DataFrame()

# ------------------------------------------------------------------
# 5️⃣  Remove deleted events
# ------------------------------------------------------------------
def purge_deleted_events(df: pd.DataFrame) -> pd.DataFrame:
    """Delete rows that match events reported as deleted."""
    if last_execution_time is None:
        # First run → nothing to purge
        return df

    page_size = 1000
    variables = {
        "pageSize": page_size,
        "polygonTypes": POLYGON_TYPES,
        "deletedAt": {"from": last_execution_time},
    }

    query = """
    query deletedPolygonEvents(
        $pageSize: Int,
        $afterCursor: String,
        $polygonTypes: [String],
        $deletedAt: RangeDate
    ) {
        deletedPolygonEvents(
            first: $pageSize,
            after: $afterCursor,
            polygonTypes: $polygonTypes,
            deletedAt: $deletedAt
        ) {
            pageInfo { endCursor }
            edges {
                node { _id }
            }
        }
    }
    """

    cursor = None
    deleted_ids = set()
    while True:
        variables["afterCursor"] = cursor
        payload = gql_query(query, variables)
        edges = payload["deletedPolygonEvents"]["edges"]
        deleted_ids.update(e["node"]["_id"] for e in edges)

        if len(edges) < page_size:
            break
        cursor = payload["deletedPolygonEvents"]["pageInfo"]["endCursor"]

    if deleted_ids:
        df = df[~df["_id"].isin(deleted_ids)].reset_index(drop=True)
    return df

# ------------------------------------------------------------------
# 6️⃣  Main integration
# ------------------------------------------------------------------
def main():
    logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")

    # 6.1 Fetch new/updated events
    new_events = fetch_polygon_events()
    logging.info(f"Fetched {len(new_events)} new/updated events")

    # 6.2 Merge with existing cache
    if not data.empty:
        data = pd.concat([data, new_events], ignore_index=True)
        data = data.drop_duplicates(subset=["_id"], keep="last")
    else:
        data = new_events

    # 6.3 Remove events that were deleted
    data = purge_deleted_events(data)
    logging.info(f"Dataset after purge: {len(data)} records")

    # 6.4 Persist
    data.to_csv(CSV_FILE, index=False)
    logging.info(f"Persisted {CSV_FILE}")

if __name__ == "__main__":
    main()

Implementation notes

pandas.json_normalize flattens the nested GraphQL structure into a tidy dataframe.
The lastUpdated filter is inclusive; subtracting 24 h from the file’s mtime guarantees that we don’t miss updates on the boundary day.
pageSize can be tuned to match network capacity and available memory.

5. Best Practices

Practice

Why

Secure token	Store it in a `.env` file or environment variable.
Limit `pageSize`	Prevent server/client time‑outs.
Logging	Enable debugging output (`logging`).
Error handling	Catch HTTP errors (`requests.exceptions.HTTPError`) and retry with exponential back‑off.
Unit tests	Mock API responses to validate merge and purge logic.
Monitoring	Run via Cron/Argo‑workflow and track record counts per run.

6. Advanced Use‑Cases

Scenario

Solution

Sync to SQL	Use `SQLAlchemy` to bulk‑load the dataframe.
Partitioning	Separate CSVs or tables by year/month.
Integrity check	Store a SHA-256 hash of the payload and verify it each time.

7. Conclusion

This approach lets you:

Significantly cut down network traffic (only changed data).
Keep your local cache fresh (add, update, delete).
Easily automate the process with a clear, maintainable script.

By following this guide, you can integrate AXSMarine’s polygon events into your document‑management ecosystem while keeping bandwidth and storage usage minimal. 🚀

References

Sections