Optimization of Tinode Message Processing in a Customer Service Dashboard

The article analyzes performance bottlenecks in a customer‑service workbench that integrates online, phone, ticket and tool modules via iframes and relies heavily on the third‑party Tinode SDK. Message reception and sending involve extensive loops for deserialization, sorting, deduplication and state handling, which become severe when the broadcast communication mode is used.

To address these issues, the author proposes several optimizations:

Deduplication: Maintain a global msgidCacheMaps Map indexed by sessionId and msgid to record processed messages and skip duplicates.

Sorting: Use a binary‑search insertion sort backed by a seqCacheMaps Map (sessionId → seq) to place messages efficiently, handling both virtual and real sequence numbers.

Cache Reclamation: Destroy the global hash maps when a session ends, is transferred, or goes offline to prevent memory overflow.

Message Status Update: Replace full‑history traversal with a binary‑search locate‑and‑update approach, updating only relevant messages (read, received, failed, etc.).

Sensitive‑Word Interception: Render messages instantly on the client, let the gateway perform keyword checks, and update the UI based on the gateway’s response without blocking the send operation.

Performance data collected before and after the deployment shows a reduction of average first‑response time from 8.40 s to 6.82 s and overall average response time from 19.9 s to 18.22 s, confirming the effectiveness of the optimizations.

The work demonstrates that careful cache design, algorithmic improvements, and asynchronous handling can significantly enhance real‑time IM performance in high‑traffic customer‑service scenarios.