Advertising Systems
Advertising Event Data: Define the Contract Before Building the Pipeline
A duplicate click illustrates the decisions behind event grain, identity, time, versions, and data-quality acceptance criteria.
Posts
Advertising Systems
Ad Attribution and Reconciliation: Why Two Correct Reports Can Disagree
A conversion spanning several days explains event facts, attribution results, metric definitions, and systematic reconciliation.
Advertising Systems
Heterogeneous Ad Data Ingestion: Configuration Still Needs a Release Process
A shared ingestion core for JSON, CSV, and Protobuf, with explicit boundaries for dynamic mapping, replay, and staged releases.
Flink
Real-Time Deduplication in Flink: State, Redis, or Bitmap?
Compare deduplication scope, recovery semantics, capacity, and the failure windows hidden inside three common approaches.
Flink
Upgrading Flink Jobs: Turn Parallel Runs and Diffs into Evidence
Separate state compatibility, result correctness, and performance regressions to make streaming migrations diagnosable and reversible.
Flink
Debugging Flink Backpressure and Checkpoints: Locate the Slow Stage First
Use a source-to-sink pipeline to distinguish downstream bottlenecks, data skew, state I/O, and barrier-alignment delays.
Databases & MySQL
MySQL Index Design: Reading EXPLAIN Through a Slow Query
Connect query shape, composite indexes, actual scan volume, and write costs through a practical event-log query.
Databases & MySQL
MySQL Concurrency: MVCC, Lock Waits, and Deadlocks in One Example
Use two sessions to understand snapshot reads and conflicting updates, inspect deadlocks, and design retries at the transaction boundary.
Databases & MySQL
Consistency Boundaries in MySQL CDC: Snapshots, Offsets, and Idempotency
Trace duplicates, reordering, deletes, and replay from MySQL through Debezium and Kafka to downstream state.
Java
Java Thread-Pool Capacity: Keeping Backpressure Under Control
Use worker, queue, and rejection limits to control how bursts turn into latency and memory pressure, then validate overload behavior.
Java
JVM Diagnostics: Separate CPU, Waiting, and Memory Before Tuning
Build an evidence chain from service symptoms to threads, JFR, garbage collection, and native memory in Java 17.
Java
Java Async Calls: Timeouts, Cancellation, and Retries Are Separate Decisions
Separate CompletableFuture completion from the underlying request lifecycle, then constrain retries with deadlines, idempotency, and explicit budgets.
AI Agents
Operations Agents: Separate Diagnosis, Approval, and Execution
Design an auditable workflow for incident evidence, action plans, idempotent execution, and rollback verification.
AI Agents
Data Agents: Executable SQL Is Only the Start of a Trustworthy Answer
Use an advertising analytics example to connect query permissions, metric definitions, scan budgets, and result validation.
AI Agents
Multi-Agent Orchestration: Design Handoffs and Recovery Before Adding Roles
Design task boundaries, evidence contracts, conflict resolution, idempotent recovery, and budgets for a data incident investigation.
AI Agents
Agent Evaluation and Replay: Iterate on Verifiable Outcomes
Build task-level evaluations for operations and data agents, separating answers, actions, and outcomes while versioning replay environments and graders.
Earlier notes · 2023
GeoMesa and HBase — archived entry
This page preserves the original URL of an early notebook entry. Continue with the engineering series for the current articles.
Earlier notes · 2023
MySQL Study Notes: Locks, Lock Ranges, and Deadlocks
An English reading edition of the 2023 notes on global, table, metadata, intention, auto-increment, and row locks, with practical inspection and recovery guidance.
Earlier notes · 2023
MySQL Study Notes: Transactions, Logging, and MVCC
An English reading edition of the 2023 notes, covering isolation, undo and redo, binary logs, commit coordination, replication, and buffer-pool behavior.
Earlier notes · 2023
Java basics — archived entry
This page preserves the original URL of an early notebook entry. Continue with the engineering series for the current articles.
Earlier notes · 2023
MySQL Fundamentals: Data Models, Query Execution, and Row Storage
An edited English edition of the archived study notes, covering relational design, storage engines, SQL execution, and InnoDB records.
Earlier notes · 2023
MySQL Indexes: B+ Trees, Composite Keys, and Execution Plans
An edited English edition of the archived indexing notes, from InnoDB pages and composite-key ranges to EXPLAIN and COUNT semantics.
Earlier notes · 2021
Merging Jupyter Notebooks and Fixing an Encoding Problem
An archived note on combining notebook cells and outputs with nbmerge, and diagnosing an incorrectly encoded output file.
Earlier notes · 2020
Traffic Flow Prediction: A Research Resource Directory
An archived collection of papers, competition solutions, PyTorch examples, and spatiotemporal forecasting resources from my earlier research.
Earlier notes · 2020
MCM/ICM Retrospective: Problem Selection, Clear Reasoning, and Visual Communication
Lessons from our 2020 Problem A Finalist submission, including a costly change of direction and the value of a coherent technical narrative.
Earlier notes · 2020
Installing TensorFlow 2.1 with GPU Support on Windows: A Historical Note
An archived Windows setup report, covering CUDA, cuDNN, the Visual C++ runtime and the distinction between a TensorFlow smoke test and GPU verification.
Earlier notes · 2020
Lessons from China's Undergraduate Mathematical Modeling Contest
A retrospective on team preparation, modeling practice, implementation, and communicating a solution under a three-day deadline.