[node]
[network]
[storage]
[cluster]
[security]
[logging]

# =============================================================================
# ThunderDB Configuration File
# =============================================================================

# -----------------------------------------------------------------------------
# Node Settings
# -----------------------------------------------------------------------------
[node]
# Unique identifier for this node in the cluster.
# Must be unique across all nodes. In Kubernetes, this is typically derived
# from the pod ordinal index.
node_id = 1

# -----------------------------------------------------------------------------
# Network Settings
# -----------------------------------------------------------------------------
[network]
# Address to bind all listeners to.
# Use "0.0.0.0" to listen on all interfaces, or a specific IP to restrict.
listen_addr = "0.0.0.0"

# PostgreSQL wire protocol port.
# Compatible with psql, pgcli, and all PostgreSQL client libraries.
pg_port = 5432

# MySQL wire protocol port.
# Compatible with mysql CLI and all MySQL client libraries.
mysql_port = 3306

# RESP (Redis Serialization Protocol) port.
# Compatible with redis-cli and all Redis client libraries.
resp_port = 6379

# HTTP API port.
# Used for REST API, admin endpoints, metrics, and health checks.
http_port = 8088

# gRPC port.
# Used for inter-node communication and the native gRPC client API.
grpc_port = 9090

# -----------------------------------------------------------------------------
# Storage Settings
# -----------------------------------------------------------------------------
[storage]
# Directory for storing data files (pages, indexes, metadata).
data_dir = "/var/lib/thunderdb/data"

# Directory for write-ahead log (WAL) files.
# For best performance, place on a separate disk from data_dir.
wal_dir = "/var/lib/thunderdb/wal"

# Size of the buffer pool (in-memory page cache).
# This is the single most important tuning parameter. Larger values improve
# read performance by keeping more pages in memory.
# Supports suffixes: KB, MB, GB.
buffer_pool_size = "128MB"

# Size of the WAL write buffer.
# Larger values improve write throughput by batching WAL writes.
wal_buffer_size = "16MB"

# Size of each data page.
# Changing this after initialization requires a full data migration.
# Valid values: 4KB, 8KB, 16KB, 32KB.
page_size = "16KB"

# Interval between automatic checkpoints.
# Checkpoints flush dirty pages to disk, reducing recovery time.
# Lower values reduce recovery time but increase I/O.
checkpoint_interval = "60s"

# Number of threads dedicated to background compaction.
# More threads speed up compaction but consume CPU.
compaction_threads = 2

# Enable direct I/O to bypass the OS page cache.
# Recommended for production to avoid double-caching.
direct_io = false

# Enable compression for data pages on disk.
# Reduces storage requirements at a small CPU cost.
compression = true

# Compression algorithm to use when compression is enabled.
# Options: "Lz4" (fast), "Snappy" (balanced), "Zstd" (high ratio).
compression_algorithm = "Lz4"

# Maximum WAL size before forcing a checkpoint.
# When the WAL reaches this size, a checkpoint is triggered regardless
# of the checkpoint_interval.
max_wal_size = "1GB"

# Whether to flush WAL to disk on every commit.
# true:  Guarantees durability (no data loss on crash). Recommended for production.
# false: Better write performance but risks losing the last few transactions on crash.
sync_commit = true

# -----------------------------------------------------------------------------
# Cluster Settings
# -----------------------------------------------------------------------------
[cluster]
# Cluster name. All nodes in the same cluster must use the same name.
cluster_name = "default"

# List of peer node addresses (host:grpc_port).
# Exclude the current node's address.
peers = []

# Raft election timeout.
# If a follower doesn't hear from the leader within this duration,
# it starts a new election. Must be greater than raft_heartbeat_interval.
# For WAN deployments, increase to 3-5s.
raft_election_timeout = "1s"

# Raft heartbeat interval.
# The leader sends heartbeats at this interval.
# Must be significantly less than raft_election_timeout (typically 1/10th).
raft_heartbeat_interval = "100ms"

# Number of copies of each data region.
# 3 is recommended for production (tolerates 1 node failure).
# Cannot exceed the number of nodes in the cluster.
replication_factor = 3

# Maximum size of a single data region before it is split.
# Smaller regions enable finer-grained load balancing.
max_region_size = "256MB"

# Minimum size of a single data region before it is merged.
# Prevents excessive fragmentation from many small regions.
min_region_size = "64MB"

# Enable automatic region balancing across nodes.
# When enabled, the leader periodically rebalances regions
# to maintain even distribution.
auto_balance = true

# -----------------------------------------------------------------------------
# Security Settings
# -----------------------------------------------------------------------------
[security]
# Enable client authentication.
# When false, all connections are accepted without credentials.
authentication_enabled = false

# Enable TLS for all client-facing protocols.
tls_enabled = false

# Path to the TLS certificate file (PEM format).
tls_cert_path = ""

# Path to the TLS private key file (PEM format).
tls_key_path = ""

# Superuser account name.
superuser = "admin"

# Superuser password (plaintext, for initial setup only).
# In production, use THUNDERDB_SUPERUSER_PASSWORD_HASH environment variable
# with an Argon2 hash instead.
superuser_password = ""

# -----------------------------------------------------------------------------
# Logging Settings
# -----------------------------------------------------------------------------
[logging]
# Log level. Options: "trace", "debug", "info", "warn", "error".
# Use "info" for production, "debug" for development, "trace" for deep debugging.
level = "info"

# Log output format. Options: "text" (human-readable), "json" (structured).
# Use "json" for production environments with log aggregation.
format = "text"

# Enable slow query logging.
# Queries exceeding the threshold are logged at WARN level.
slow_query_enabled = true

# Threshold for slow query logging.
# Queries taking longer than this are logged.
slow_query_threshold = "1s"

docker run -d \
  -e THUNDERDB_NODE_ID=1 \
  -e THUNDERDB_LISTEN_ADDR=0.0.0.0 \
  -e THUNDERDB_LOG_LEVEL=info \
  -e THUNDERDB_SUPERUSER_PASSWORD_HASH='argon2:$argon2id$v=19$m=65536,t=3,p=4$...' \
  -e THUNDERDB_DATA_DIR=/var/lib/thunderdb/data \
  -e THUNDERDB_WAL_DIR=/var/lib/thunderdb/wal \
  thunderdb:latest

sudo systemctl edit thunderdb

[Service]
Environment="THUNDERDB_LOG_LEVEL=debug"
Environment="THUNDERDB_SUPERUSER_PASSWORD_HASH=argon2:$argon2id$v=19$..."

[storage]
# Large buffer pool to keep hot rows in memory.
# Aim for 60-70% of available system RAM.
buffer_pool_size = "8GB"

# Moderate WAL buffer -- OLTP writes are typically small.
wal_buffer_size = "32MB"

# Ensure every commit is durable.
sync_commit = true

# Frequent checkpoints reduce recovery time after crashes.
checkpoint_interval = "30s"

# Fewer compaction threads needed; OLTP generates less bulk data.
compaction_threads = 2

# Bypass OS cache to avoid double-buffering.
direct_io = true

# Lz4 for minimal CPU overhead on the write path.
compression = true
compression_algorithm = "Lz4"

# Smaller max WAL keeps recovery time bounded.
max_wal_size = "512MB"

[storage]
# Moderate buffer pool -- OLAP scans are sequential and don't benefit
# as much from caching random pages.
buffer_pool_size = "4GB"

# Large WAL buffer to handle bulk writes efficiently.
wal_buffer_size = "128MB"

# Async commit is acceptable if some data loss on crash is tolerable.
sync_commit = false

# Less frequent checkpoints to reduce I/O during long-running queries.
checkpoint_interval = "300s"

# More compaction threads for faster background processing of bulk data.
compaction_threads = 8

# Direct I/O is still beneficial for large sequential reads.
direct_io = true

# Zstd compression for maximum space savings on large datasets.
compression = true
compression_algorithm = "Zstd"

# Larger max WAL to avoid checkpoint storms during bulk loads.
max_wal_size = "4GB"

[storage]
# Balance between caching hot transactional data and leaving room
# for analytical query memory needs.
buffer_pool_size = "6GB"

# Balanced WAL buffer.
wal_buffer_size = "64MB"

# Durability is important for the transactional component.
sync_commit = true

# Moderate checkpoint interval balances recovery time and I/O.
checkpoint_interval = "60s"

# Moderate compaction thread count.
compaction_threads = 4

# Direct I/O recommended.
direct_io = true

# Lz4 is a good default balance of speed and compression.
compression = true
compression_algorithm = "Lz4"

# Moderate max WAL size.
max_wal_size = "1GB"

# Increase file descriptor limits
echo "thunder soft nofile 65535" >> /etc/security/limits.conf
echo "thunder hard nofile 65535" >> /etc/security/limits.conf

# Reduce swappiness (prefer keeping data in RAM)
echo "vm.swappiness=1" >> /etc/sysctl.conf

# Increase the number of memory map areas
echo "vm.max_map_count=262144" >> /etc/sysctl.conf

# Increase network buffer sizes for high-throughput workloads
echo "net.core.somaxconn=65535" >> /etc/sysctl.conf
echo "net.ipv4.tcp_max_syn_backlog=65535" >> /etc/sysctl.conf

# Apply changes
sysctl -p

# Use noop or none scheduler for SSDs
echo "none" > /sys/block/sda/queue/scheduler

# Set readahead to 256 KB for SSDs (reduce for random I/O workloads)
blockdev --setra 512 /dev/sda