Design and Implementation of QQ Reminder Subscription Push System

QQ serves a massive mobile user base, and its reminder subscription feature is critical for both users and business partners, requiring high reliability, controllable rate, and efficient delivery.

The solution adopts a Dispatcher + Worker model with a hybrid push‑pull mechanism: tasks are pre‑downloaded to clients at 00:00 (pull) while remaining notifications are pushed, ensuring timely reminders and preventing last‑minute modifications.

Data is stored heterogeneously: task metadata resides in MySQL for durability, while the massive subscription list (user UINs) is kept in Redis sets for fast read/write, leveraging Redis's single‑threaded model, set‑based de‑duplication, and efficient bulk spop operations.

To guarantee timely execution, a distributed timer based on Redis sorted sets handles delayed tasks, complemented by a local database poller that retries after a 3‑second fallback, ensuring idempotent execution.

Controllable scheduling is achieved with a distributed rate limiter using Redis INCR to count per‑second dispatches; if the limit is exceeded, dispatchers wait for the next second. Pseudocode:

CREATE TABLE table_xxx(
    ds BIGINT COMMENT '数据日期',
    label_name STRING COMMENT '标签名称',
    label_id BIGINT COMMENT '标签id',
    appid STRING COMMENT '小程序appid',
    useruin BIGINT COMMENT 'useruin',
    tag_name STRING COMMENT 'tag名称',
    tag_id BIGINT COMMENT 'tag id',
    tag_value BIGINT COMMENT 'tag 权重值'
) PARTITION BY LIST(ds) SUBPARTITION BY LIST(label_name) (
    SUBPARTITION sp_xxx VALUES IN ('xxx'),
    SUBPARTITION sp_xxxx VALUES IN ('xxxx')
);

Large tasks are split into 5,000‑UIN chunks, allowing fine‑grained rate control; workers fetch each chunk via spop and push notifications concurrently.

A message queue decouples scheduling from execution, providing peak‑shaving, asynchronous processing, and at‑least‑once delivery guarantees.

For at‑least‑once push, a Redis Lua script atomically moves a batch of UINs from the subscription set to a task set, preserving the data if the push fails:

redis.replicate_commands()
local set_key, task_key = KEYS[1], KEYS[2]
local num = tonumber(ARGV[1])
local array = redis.call('SPOP', set_key, num)
if #array > 0 then
    redis.call('SADD', task_key, unpack(array))
end
return redis.call('SCARD', task_key)

Disaster recovery uses dual storage: each subscription event is recorded in Tencent's CKV KV‑store and mirrored in a Redis cluster (2 replicas, 3 shards), enabling recovery from either source.

Overall, the Dispatcher + Worker architecture, combined with hybrid push‑pull, heterogeneous storage, distributed rate limiting, task chunking, and reliable messaging, delivers a scalable, controllable, and fault‑tolerant push notification service for QQ.