Exercises

# ============================================================
# Drill 1 — Last-seen for online presence
# ============================================================
# TASK: track 'last seen at' for 100k users; query 'is X online (seen in last 60s)'.
#
# ANSWER: SORTED SET keyed on user id, score = epoch ms.
#   ZADD presence $now user_id
#   Query active in last 60s: ZRANGEBYSCORE presence ($now - 60_000) +inf
#   Garbage collect with: ZREMRANGEBYSCORE presence 0 ($now - 60_000)
# Why: O(log N) writes, O(log N + M) reads, automatic 'sorted by time'.
# Memory: ~50 MB at 100k users — well within budget.

# ============================================================
# Drill 2 — Cooldown timer per action
# ============================================================
# TASK: 'user can claim daily reward at most once per 24h'.
#
# ANSWER: SET ... NX EX
#   SET cooldown:claim:42 1 NX EX 86400
#   On nil response, reject (the lock was held).
# Why: atomic claim + TTL in one round trip.

# ============================================================
# Drill 3 — Top-N visited pages this minute
# ============================================================
# TASK: count page views per URL, return top 10 for the current minute.
#
# ANSWER: ZSET per minute bucket.
#   ZINCRBY views:202606181232 1 /url
#   ZREVRANGE views:202606181232 0 9 WITHSCORES
#   EXPIRE views:202606181232 600    # keep only last 10 buckets
# Why: O(log N) increments, O(log N + 10) top read.

# ============================================================
# Drill 4 — Dedup messages across a moving 1-hour window
# ============================================================
# TASK: incoming webhook ids must be processed AT MOST ONCE per hour.
#
# ANSWER: SET ... NX EX
#   SET dedupe:$id 1 NX EX 3600
# Treat non-OK reply as 'duplicate, skip'.

# ============================================================
# Drill 5 — Fan-out a notification to all subscribers
# ============================================================
# TASK: 100 subscribers want every 'order.shipped' event delivered.
#
# ANSWER: Streams + Consumer Groups.
#   XADD orders.shipped * order_id 42 user_id 7
#   XGROUP CREATE orders.shipped notify-svc $ MKSTREAM
#   XREADGROUP GROUP notify-svc consumer-1 COUNT 16 BLOCK 5000 STREAMS orders.shipped >
#   XACK orders.shipped notify-svc <id>
# Why: durable, supports retry via XPENDING. Pub/Sub has no replay.

# ============================================================
# Drill 6 — Job scheduler ('run after 5 minutes')
# ============================================================
# TASK: schedule jobs to run at future timestamps; consumer pulls due jobs.
#
# ANSWER: ZSET keyed by run_at score.
#   ZADD jobs $runAt $jobId
#   Consumer loop:
#     ZRANGEBYSCORE jobs 0 $now LIMIT 0 16
#     for each id: ZREM jobs id; process(id)
# Why: O(log N) writes, O(log N + M) reads of due items.

# ============================================================
# Scoring
# ============================================================
# 6 / 6 -> ready for production caches and queues
# 4 / 6 -> revisit redis/cheatsheet
# < 4   -> read https://redis.io/docs/data-types/

# ============================================================
# Pitfalls
# ============================================================
# - 'PUBSUB for job queues' -> no replay; consumers offline = events lost
# - 'KEYS *' to enumerate -> O(N) blocking; use SCAN
# - Storing 1MB JSON blobs in strings -> Redis is in-memory; budget bites
# - No maxmemory + eviction policy -> OOM kills the server

# Fill in: ____ counter      # increment by 1