Building a PHP Extension for Automated Web API Monitoring at 58 Anjuke

The article introduces a PHP‑extension‑based solution for automating Web API monitoring, aiming to simplify API health checks for 58 Anjuke's PHP services.

Background : Rapid growth of rental business APIs led to a lack of unified monitoring tools, causing issues such as timeout detection, traffic analysis, and performance insight.

Design Principles :

Flexible customization: standardize collected data (cluster name, URI, HTTP code, latency, etc.) and allow configurable collection intervals and URI models.

Low cost integration: load the extension at PHP process start, requiring no code changes from developers.

High resource utilization: use shared memory and thread‑safe mechanisms to minimize CPU and memory impact.

Configuration example:

wrm.enable=1
wrm.tick=60
wrm.timeout=2000
wrm.collector_register_url=https://****.**.com/****/****
wrm.collector_port=****
wrm.cluster=***_***_***
wrm.total_url_num=500
wrm.enable_sync_uri=1
wrm.sync_uri_url=****.**.com/****/****

System Architecture :

Process startup: PHP Zend engine loads the dynamic extension, forks a monitoring process via the MINIT hook.

Log collection: RINIT gathers request data (URI, HTTP code, latency) into a trace structure; RSHUTDOWN records response metrics.

Data reporting: a timer‑driven thread swaps hash tables, compresses data, and asynchronously reports to the central service.

Key technical challenges addressed:

Preventing reporting traffic spikes by aggregating logs locally before periodic compressed uploads.

Designing an efficient hash‑based data structure using the bkrdhash algorithm to store per‑URI metrics within ~10 MB for 10 000 URIs.

Ensuring concurrent writes with futex kernel‑level locks.

Isolating collection and reporting via dual shared‑memory regions (A and B) for thread‑safe asynchronous reporting.

Low‑cost deployment: the .so extension works with PHP 7+ and can be loaded via php.ini or project‑level extension directories.

Results : Deployed across dozens of clusters, covering over a hundred PHP servers, collecting >1 million logs per minute, achieving near‑real‑time monitoring, reducing incident response from days to minutes, and providing detailed traffic analytics for capacity planning.

Authors : Chen Yitian and Hu Kun from 58 Rental Technology Department.