Category: Linux OS

Application A  ←→  Socket  ←→  Network  ←→  Socket  ←→  Application B

# Two Ruby programs talking via sockets
# Program 1 (Server)
require 'socket'
server = TCPServer.new(3000)
client_socket = server.accept
client_socket.puts "Hello from server!"

# Program 2 (Client)  
client = TCPSocket.new('localhost', 3000)
message = client.gets
puts message  # "Hello from server!"

# Talk to Redis (what Sidekiq does)
require 'socket'
redis_socket = TCPSocket.new('localhost', 6379)
redis_socket.write("PING\r\n")
response = redis_socket.read  # "PONG"

# Talk across the internet
require 'socket'
socket = TCPSocket.new('google.com', 80)
socket.write("GET / HTTP/1.1\r\nHost: google.com\r\n\r\n")

# Talk between programs on same machine
# Faster than network sockets - no network stack overhead
socket = UNIXSocket.new('/tmp/my_app.sock')

# Network socket (goes through TCP/IP stack)
redis = Redis.new(host: 'localhost', port: 6379)

# Unix socket (direct OS communication)
redis = Redis.new(path: '/tmp/redis.sock')  # Faster!

IP Address: 192.168.1.100 (Building address)
Port: 6379                (Apartment number)

Same computer running:
- Web server on port 80
- Redis on port 6379  
- SSH on port 22
- Your app on port 3000

When data arrives at 192.168.1.100:6379
→ OS knows to send it to Redis

# Different services need different "apartments"
$ netstat -ln
tcp 0.0.0.0:22    SSH server
tcp 0.0.0.0:80    Web server  
tcp 0.0.0.0:443   HTTPS server
tcp 0.0.0.0:3306  MySQL
tcp 0.0.0.0:5432  PostgreSQL
tcp 0.0.0.0:6379  Redis
tcp 0.0.0.0:27017 MongoDB

Redis server (port 6379) can handle:
- Connection 1: Sidekiq worker
- Connection 2: Rails app  
- Connection 3: Redis CLI
- Connection 4: Monitoring tool

Each gets a unique "channel" but all use port 6379

0-1023:    Reserved (HTTP=80, SSH=22, etc.)
1024-49151: Registered applications  
49152-65535: Dynamic/Private (temporary connections)

# What happens when you do this:
socket = TCPSocket.new('localhost', 6379)

// OS system calls
socket_fd = socket(AF_INET, SOCK_STREAM, 0)  // Create socket
connect(socket_fd, server_address, address_len)  // Connect

Process ID: 1234 (Your Ruby app)
File Descriptors:
  0: stdin
  1: stdout  
  2: stderr
  3: socket to Redis (localhost:6379)
  4: socket to PostgreSQL (localhost:5432)
  5: listening socket (port 3000)

Application: socket.write("PING")
     ↓
Ruby: calls OS write() system call
     ↓  
Kernel: adds to socket send buffer
     ↓
Network Stack: TCP → IP → Ethernet
     ↓
Network Card: sends packets over wire

# Like registered mail - guaranteed delivery
server = TCPServer.new(3000)
client = TCPSocket.new('localhost', 3000)

# Data arrives in order, no loss
client.write("Message 1")
client.write("Message 2") 
# Server receives exactly: "Message 1", "Message 2"

# Like shouting across a crowded room
require 'socket'

# Sender
udp = UDPSocket.new
udp.send("Hello!", 0, 'localhost', 3000)

# Receiver  
udp = UDPSocket.new
udp.bind('localhost', 3000)
data = udp.recv(1024)  # Might not arrive!

# Super fast local communication
File.delete('/tmp/test.sock') if File.exist?('/tmp/test.sock')

# Server
server = UNIXServer.new('/tmp/test.sock')
# Client
client = UNIXSocket.new('/tmp/test.sock')

# 1. Server: "I'm listening on port 3000"
server = TCPServer.new(3000)

# 2. Client: "I want to connect to port 3000"  
client = TCPSocket.new('localhost', 3000)

# 3. Server: "I accept your connection"
connection = server.accept

# 4. Both can now send/receive data
connection.puts "Hello!"
client.puts "Hi back!"

# 5. Clean shutdown
client.close
connection.close
server.close

Client                    Server
  |                         |
  |---- SYN packet -------->| (I want to connect)
  |<-- SYN-ACK packet ------| (OK, let's connect)  
  |---- ACK packet -------->| (Connection established!)
  |                         |
  |<---- Data exchange ---->|
  |                         |

socket = TCPSocket.new('localhost', 6379)
puts socket.fileno  # e.g., 7

# This socket is just file descriptor #7!
# You can even use it with raw system calls

Application Buffer  ←→  Kernel Send Buffer  ←→  Network
                   ←→  Kernel Recv Buffer  ←→

socket.write("BRPOP queue 0")
# 1. Ruby copies data to kernel send buffer
# 2. write() returns immediately  
# 3. Kernel sends data in background
# 4. TCP handles retransmission, etc.

data = socket.read  
# 1. Check kernel receive buffer
# 2. If empty, BLOCK thread until data arrives
# 3. Copy data from kernel to Ruby
# 4. Return to your program

# When Sidekiq does this:
redis.brpop("queue:default", timeout: 2)

# Here's the socket journey:
# 1. Ruby opens TCP socket to localhost:6379
socket = TCPSocket.new('localhost', 6379)

# 2. Format Redis command
command = "*4\r\n$5\r\nBRPOP\r\n$13\r\nqueue:default\r\n$1\r\n2\r\n"

# 3. Write to socket (goes to kernel buffer)
socket.write(command)

# 4. Thread blocks reading response
response = socket.read  # BLOCKS HERE until Redis responds

# 5. Redis eventually sends back data
# 6. Kernel receives packets, assembles them
# 7. socket.read returns with the job data

# Bad: New socket for each request
100.times do
  socket = TCPSocket.new('localhost', 6379)
  socket.write("PING\r\n")
  socket.read
  socket.close  # Expensive!
end

# Good: Reuse socket
socket = TCPSocket.new('localhost', 6379)
100.times do
  socket.write("PING\r\n")  
  socket.read
end
socket.close

# What Redis gem/Sidekiq does internally
class ConnectionPool
  def initialize(size: 5)
    @pool = size.times.map { TCPSocket.new('localhost', 6379) }
  end

  def with_connection(&block)
    socket = @pool.pop
    yield(socket)
  ensure
    @pool.push(socket)
  end
end

# On Linux/Mac, sockets appear as files!
$ lsof -p #{Process.pid}
ruby 1234 user 3u sock 0,9 0t0 TCP localhost:3000->localhost:6379

# Your OS has limits
$ ulimit -n
1024  # Max file descriptors (including sockets)

# Web servers need thousands of sockets
# That's why they increase this limit!

$ netstat -an | grep 6379
tcp4 0 0 127.0.0.1.6379 127.0.0.1.50123 ESTABLISHED
tcp4 0 0 127.0.0.1.6379 127.0.0.1.50124 TIME_WAIT
tcp4 0 0 *.6379         *.*            LISTEN

# Traditional database thinking
User.where(active: true).pluck(:id)

# Redis thinking  
redis.smembers("active_users")  # A set of active user IDs

┌─────────────────────────────────┐
│          Client Connections     │ ← Your Ruby app connects here
├─────────────────────────────────┤
│         Command Processing      │ ← Parses your BRPOP command
├─────────────────────────────────┤
│         Event Loop (epoll)      │ ← Handles thousands of connections
├─────────────────────────────────┤
│        Data Structure Engine    │ ← The magic happens here
├─────────────────────────────────┤
│         Memory Management       │ ← Keeps everything in RAM
├─────────────────────────────────┤
│        Persistence Layer        │ ← Optional disk storage
└─────────────────────────────────┘

// Simplified Redis main loop
while (server_running) {
    // 1. Check for new network events
    events = epoll_wait(eventfd, events, max_events, timeout);

    // 2. Process each event
    for (int i = 0; i < events; i++) {
        if (events[i].type == READ_EVENT) {
            process_client_command(events[i].client);
        }
    }

    // 3. Handle time-based events (expiry, etc.)
    process_time_events();
}

// Redis list structure (simplified)
typedef struct list {
    listNode *head;      // First item
    listNode *tail;      // Last item  
    long length;         // How many items
    // ... other fields
} list;

typedef struct listNode {
    struct listNode *prev;
    struct listNode *next;
    void *value;         // Your job data
} listNode;

// Simplified BRPOP command handler
void brpopCommand(client *c) {
    // Try to pop from each list
    for (int i = 1; i < c->argc - 1; i++) {
        robj *key = c->argv[i];
        robj *list = lookupKeyRead(c->db, key);

        if (list && listTypeLength(list) > 0) {
            // Found item! Pop and return immediately
            robj *value = listTypePop(list, LIST_TAIL);
            addReplyMultiBulkLen(c, 2);
            addReplyBulk(c, key);
            addReplyBulk(c, value);
            return;
        }
    }

    // No items found - BLOCK the client
    blockForKeys(c, c->argv + 1, c->argc - 2, timeout);
}

// Redis hash table
typedef struct dict {
    dictEntry **table;       // Array of buckets
    unsigned long size;      // Size of table
    unsigned long sizemask;  // size - 1 (for fast modulo)
    unsigned long used;      // Number of entries
} dict;

// Finding a key
unsigned int hash = dictGenHashFunction(key);
unsigned int idx = hash & dict->sizemask;
dictEntry *entry = dict->table[idx];

// Simplified event loop
int epollfd = epoll_create(1024);

// Add client socket to epoll
struct epoll_event ev;
ev.events = EPOLLIN;  // Watch for incoming data
ev.data.ptr = client;
epoll_ctl(epollfd, EPOLL_CTL_ADD, client->fd, &ev);

// Main loop
while (1) {
    int nfds = epoll_wait(epollfd, events, MAX_EVENTS, timeout);

    for (int i = 0; i < nfds; i++) {
        client *c = (client*)events[i].data.ptr;

        if (events[i].events & EPOLLIN) {
            // Data available to read
            read_client_command(c);
            process_command(c);
        }
    }
}

Traditional approach: 1 thread per connection
- 1000 connections = 1000 threads
- Each thread uses ~8MB memory
- Context switching overhead

Redis approach: 1 thread for all connections  
- 1000 connections = 1 thread
- Minimal memory overhead
- No context switching between connections

// When no data available for BRPOP
typedef struct blockingState {
    dict *keys;           // Keys we're waiting for
    time_t timeout;       // When to give up
    int numreplicas;      // Replication stuff
    // ... other fields
} blockingState;

// Block a client
void blockClient(client *c, int btype) {
    c->flags |= CLIENT_BLOCKED;
    c->btype = btype;
    c->bstate = zmalloc(sizeof(blockingState));

    // Add to server's list of blocked clients
    listAddNodeTail(server.clients, c);
}

// Check for timed out clients
void handleClientsBlockedOnKeys(void) {
    time_t now = time(NULL);

    listIter li;
    listNode *ln;
    listRewind(server.clients, &li);

    while ((ln = listNext(&li)) != NULL) {
        client *c = listNodeValue(ln);

        if (c->flags & CLIENT_BLOCKED && 
            c->bstate.timeout != 0 && 
            c->bstate.timeout < now) {

            // Timeout! Send null response
            addReplyNullArray(c);
            unblockClient(c);
        }
    }
}

// When someone does LPUSH to a list
void signalKeyAsReady(redisDb *db, robj *key) {
    readyList *rl = zmalloc(sizeof(*rl));
    rl->key = key;
    rl->db = db;

    // Add to ready list
    listAddNodeTail(server.ready_keys, rl);
}

// Process ready keys and unblock clients
void handleClientsBlockedOnKeys(void) {
    while (listLength(server.ready_keys) != 0) {
        listNode *ln = listFirst(server.ready_keys);
        readyList *rl = listNodeValue(ln);

        // Find blocked clients waiting for this key
        list *clients = dictFetchValue(rl->db->blocking_keys, rl->key);

        if (clients) {
            // Unblock first client and serve the key
            client *receiver = listNodeValue(listFirst(clients));
            serveClientBlockedOnList(receiver, rl->key, rl->db);
        }

        listDelNode(server.ready_keys, ln);
    }
}

// Every Redis object has this header
typedef struct redisObject {
    unsigned type:4;        // STRING, LIST, SET, etc.
    unsigned encoding:4;    // How it's stored internally  
    unsigned lru:24;        // LRU eviction info
    int refcount;          // Reference counting
    void *ptr;             // Actual data
} robj;

// Small lists use ziplist (compressed)
if (listLength(list) < server.list_max_ziplist_entries &&
    listTotalSize(list) < server.list_max_ziplist_value) {

    // Use compressed ziplist
    listConvert(list, OBJ_ENCODING_ZIPLIST);
} else {
    // Use normal linked list
    listConvert(list, OBJ_ENCODING_LINKEDLIST);  
}

Small list: ["job1", "job2", "job3"]
Normal encoding: 3 pointers + 3 allocations = ~200 bytes
Ziplist encoding: 1 allocation = ~50 bytes (75% savings!)

// Redis memory management
void freeMemoryIfNeeded(void) {
    while (server.memory_usage > server.maxmemory) {
        // Try to free memory by:
        // 1. Expiring keys
        // 2. Evicting LRU keys  
        // 3. Running garbage collection

        if (freeOneObjectFromFreelist() == C_OK) continue;
        if (expireRandomExpiredKey() == C_OK) continue;
        if (evictExpiredKeys() == C_OK) continue;

        // Last resort: evict LRU key
        evictLRUKey();
    }
}

// Save entire dataset to disk
int rdbSave(char *filename) {
    FILE *fp = fopen(filename, "w");

    // Iterate through all databases
    for (int dbid = 0; dbid < server.dbnum; dbid++) {
        redisDb *db = server.db + dbid;
        dict *d = db->dict;

        // Save each key-value pair
        dictIterator *di = dictGetSafeIterator(d);
        dictEntry *de;

        while ((de = dictNext(di)) != NULL) {
            sds key = dictGetKey(de);
            robj *val = dictGetVal(de);

            // Write key and value to file
            rdbSaveStringObject(fp, key);
            rdbSaveObject(fp, val);
        }
    }

    fclose(fp);
}

// Log every write command
void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, 
                       robj **argv, int argc) {
    sds buf = sdsnew("");

    // Format as Redis protocol
    buf = sdscatprintf(buf, "*%d\r\n", argc);
    for (int i = 0; i < argc; i++) {
        buf = sdscatprintf(buf, "$%lu\r\n", 
                          (unsigned long)sdslen(argv[i]->ptr));
        buf = sdscatsds(buf, argv[i]->ptr);
        buf = sdscatlen(buf, "\r\n", 2);
    }

    // Write to AOF file
    write(server.aof_fd, buf, sdslen(buf));
    sdsfree(buf);
}

Operation           Operations/second
SET                 100,000+
GET                 100,000+  
LPUSH               100,000+
BRPOP (no block)    100,000+
BRPOP (blocking)    Limited by job arrival rate

# redis.conf optimizations
tcp-nodelay yes              # Disable Nagle's algorithm
tcp-keepalive 60            # Keep connections alive
timeout 0                   # Never timeout idle clients

# Memory optimizations  
maxmemory-policy allkeys-lru  # Evict least recently used
save ""                       # Disable snapshotting for speed

Master (writes) ─┐
                 ├─→ Slave 1 (reads)
                 ├─→ Slave 2 (reads)
                 └─→ Slave 3 (reads)

Client ─→ Hash Key ─→ Determine Slot ─→ Route to Correct Node

Slots 0-5460:    Node A  
Slots 5461-10922: Node B
Slots 10923-16383: Node C

# Real-time stats
redis-cli info

# Monitor all commands
redis-cli monitor

# Check slow queries
redis-cli slowlog get 10

# Memory usage by key pattern
redis-cli --bigkeys

✅ Need sub-millisecond latency
✅ Working with simple data structures  
✅ Caching frequently accessed data
✅ Session storage
✅ Real-time analytics
✅ Message queues (like Sidekiq!)

❌ Need complex queries (use PostgreSQL)
❌ Need ACID transactions across keys
❌ Dataset larger than available RAM
❌ Need strong consistency guarantees

Redis:
+ Rich data types (lists, sets, hashes)
+ Persistence options
+ Pub/sub messaging
+ Transactions
- Higher memory usage

Memcached:  
+ Lower memory overhead
+ Simpler codebase
- Only key-value storage
- No persistence

# Modern alternative to lists for job queues
redis.xadd("jobs", {"type" => "email", "user_id" => 123})
redis.xreadgroup("workers", "worker-1", "jobs", ">")

RedisJSON:     Native JSON support
RedisSearch:   Full-text search
RedisGraph:    Graph database
RedisAI:       Machine learning
TimeSeries:    Time-series data

- Multi-part AOF files
- Config rewriting improvements  
- Better memory introspection
- Enhanced security (ACLs)
- Sharded pub/sub