Class: DeadBro::SqlSubscriber
- Inherits:
-
Object
- Object
- DeadBro::SqlSubscriber
- Defined in:
- lib/dead_bro/sql_subscriber.rb
Constant Summary collapse
- SQL_EVENT_NAME =
"sql.active_record"- THREAD_LOCAL_KEY =
:dead_bro_sql_queries- THREAD_LOCAL_ALLOC_START_KEY =
:dead_bro_sql_alloc_start- THREAD_LOCAL_ALLOC_RESULTS_KEY =
:dead_bro_sql_alloc_results- THREAD_LOCAL_BACKTRACE_KEY =
:dead_bro_sql_backtraces- THREAD_LOCAL_EXPLAIN_PENDING_KEY =
:dead_bro_explain_pending- MAX_TRACKED_QUERIES =
1000- SENSITIVE_KV_QUOTED_RE =
Precompiled regexes used by sanitize_sql. Dynamic /…/i literals inside a hot-path method allocate a fresh Regexp on every call — pinning them here removes that allocation entirely.
/\b(password|token|secret|key|ssn|credit_card)\s*=\s*['"][^'"]*['"]/i- SENSITIVE_KV_BARE_RE =
/\b(password|token|secret|key|ssn|credit_card)\s*=\s*[^'",\s)]+/i- WHERE_EQ_QUOTED_RE =
/WHERE\s+[^=]+=\s*['"][^'"]*['"]/i- WHERE_EQ_QUOTED_INNER_RE =
/=\s*['"][^'"]*['"]/- SANITIZE_MAX_LENGTH =
1000- SANITIZE_SKIP_SENSITIVE_WHEN_NO_KEYWORDS =
/password|token|secret|key|ssn|credit_card/i- SANITIZE_SKIP_WHERE_WHEN_NO_KEYWORD =
/WHERE/i- MAX_PENDING_EXPLAINS =
Upper bound on pending EXPLAIN threads per request — stops a slow-query storm from spawning unbounded background threads.
20- EXPLAIN_WAIT_TIMEOUT_SECONDS =
Overall wall-clock we’re willing to block the request thread for pending EXPLAINs. Dropped from 5s → 1s: if the plan isn’t ready by then, skip it rather than stall the request.
1.0- APP_BACKTRACE_MAX_FRAMES =
25- APP_BACKTRACE_SENSITIVE_RE =
/\/[^\/]*(password|secret|key|token)[^\/]*\//i
Class Method Summary collapse
-
.append_log_to_thread(thread, severity, message) ⇒ Object
Append a log entry directly to a specific thread’s log storage This is used when logging from background threads to ensure logs are collected with the main request thread’s logs.
- .build_explain_query(sql, connection) ⇒ Object
-
.capture_app_backtrace ⇒ Object
Cheap app-only backtrace for the current query.
-
.current_queries_array ⇒ Object
Current queries array (top of stack); nil if no active tracking.
- .format_explain_result(result, connection) ⇒ Object
- .interpolate_sql_with_binds(sql, binds, connection) ⇒ Object
- .safe_query_trace(data, captured_backtrace = nil) ⇒ Object
- .sanitize_sql(sql) ⇒ Object
-
.should_continue_tracking?(current_queries_array, max_count) ⇒ Boolean
Check if we should continue tracking based on count and time limits.
- .should_explain_query?(duration_ms, sql) ⇒ Boolean
- .start_explain_analyze_background(sql, connection_id, query_info, binds = nil) ⇒ Object
- .start_request_tracking ⇒ Object
- .stop_request_tracking ⇒ Object
- .subscribe! ⇒ Object
-
.tracking_active? ⇒ Boolean
True when there is at least one active tracking context (e.g. for nested jobs).
- .wait_for_pending_explains(timeout_seconds) ⇒ Object
Class Method Details
.append_log_to_thread(thread, severity, message) ⇒ Object
Append a log entry directly to a specific thread’s log storage This is used when logging from background threads to ensure logs are collected with the main request thread’s logs
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 |
# File 'lib/dead_bro/sql_subscriber.rb', line 272 def self.append_log_to_thread(thread, severity, ) = Time.now.utc log_entry = { sev: severity.to_s, msg: .to_s, time: .iso8601(3) } # Append to the specified thread's log storage thread[:dead_bro_logs] ||= [] thread[:dead_bro_logs] << log_entry # Also print the message immediately (using current thread's logger) begin if defined?(Rails) && Rails.respond_to?(:logger) && Rails.logger = "[DeadBro] #{.iso8601(3)} #{severity.to_s.upcase}: #{}" case severity when :debug Rails.logger.debug() when :info Rails.logger.info() when :warn Rails.logger.warn() when :error Rails.logger.error() when :fatal Rails.logger.fatal() end else # Fallback to stdout $stdout.puts("[DeadBro] #{.iso8601(3)} #{severity.to_s.upcase}: #{}") end rescue # Never let logging break the application $stdout.puts("[DeadBro] #{severity.to_s.upcase}: #{}") end end |
.build_explain_query(sql, connection) ⇒ Object
310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 |
# File 'lib/dead_bro/sql_subscriber.rb', line 310 def self.build_explain_query(sql, connection) adapter_name = connection.adapter_name.downcase case adapter_name when "postgresql", "postgis" # PostgreSQL supports ANALYZE and BUFFERS "EXPLAIN (ANALYZE, BUFFERS) #{sql}" when "mysql", "mysql2", "trilogy" # MySQL uses different syntax - ANALYZE is a separate keyword "EXPLAIN ANALYZE #{sql}" when "sqlite3" # SQLite supports EXPLAIN QUERY PLAN "EXPLAIN QUERY PLAN #{sql}" else # Generic fallback - just EXPLAIN "EXPLAIN #{sql}" end end |
.capture_app_backtrace ⇒ Object
Cheap app-only backtrace for the current query. Uses caller_locations (lazy frame objects, no string allocations until we render) and keeps only frames under app/ (filtering vendor/). Returns at most N frames.
445 446 447 448 449 450 451 452 453 454 455 456 457 458 |
# File 'lib/dead_bro/sql_subscriber.rb', line 445 def self.capture_app_backtrace locations = caller_locations(1, 100) || [] frames = [] locations.each do |loc| path = loc.path.to_s next unless path.include?("app/") next if path.include?("/vendor/") frames << "#{path}:#{loc.lineno}:in `#{loc.label}'".gsub(APP_BACKTRACE_SENSITIVE_RE, "/[FILTERED]/") break if frames.length >= APP_BACKTRACE_MAX_FRAMES end frames rescue [] end |
.current_queries_array ⇒ Object
Current queries array (top of stack); nil if no active tracking.
37 38 39 40 41 |
# File 'lib/dead_bro/sql_subscriber.rb', line 37 def self.current_queries_array stack = Thread.current[THREAD_LOCAL_KEY] return nil unless stack.is_a?(Array) && stack.any? stack.last end |
.format_explain_result(result, connection) ⇒ Object
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 |
# File 'lib/dead_bro/sql_subscriber.rb', line 369 def self.format_explain_result(result, connection) adapter_name = connection.adapter_name.downcase case adapter_name when "postgresql", "postgis" # PostgreSQL returns ActiveRecord::Result from select_all if result.respond_to?(:rows) # ActiveRecord::Result object - rows is an array of arrays # Each row is [query_plan_string] plan_text = result.rows.map { |row| row.is_a?(Array) ? row.first.to_s : row.to_s }.join("\n") return plan_text unless plan_text.strip.empty? end # Try alternative methods to extract the plan if result.respond_to?(:each) && result.respond_to?(:columns) # ActiveRecord::Result with columns plan_column = result.columns.find { |col| col.downcase.include?("plan") || col.downcase.include?("query") } || result.columns.first plan_text = result.map { |row| if row.is_a?(Hash) row[plan_column] || row[plan_column.to_sym] || row.values.first else row end }.join("\n") return plan_text unless plan_text.strip.empty? end if result.is_a?(Array) # Array of hashes or arrays plan_text = result.map do |row| if row.is_a?(Hash) row["QUERY PLAN"] || row["query plan"] || row[:query_plan] || row.values.first.to_s elsif row.is_a?(Array) row.first.to_s else row.to_s end end.join("\n") return plan_text unless plan_text.strip.empty? end # Fallback to string representation result.to_s when "mysql", "mysql2", "trilogy" # MySQL returns Mysql2::Result object which needs to be converted to array if result.respond_to?(:to_a) # Convert Mysql2::Result to array of hashes rows = result.to_a rows.map { |row| row.is_a?(Hash) ? row.values.join(" | ") : row.to_s }.join("\n") elsif result.is_a?(Array) result.map { |row| row.is_a?(Hash) ? row.values.join(" | ") : row.to_s }.join("\n") else result.to_s end when "sqlite3" # SQLite returns rows if result.is_a?(Array) result.map { |row| row.is_a?(Hash) ? row.values.join(" | ") : row.to_s }.join("\n") else result.to_s end else # Generic fallback result.to_s end rescue # Fallback to string representation result.to_s end |
.interpolate_sql_with_binds(sql, binds, connection) ⇒ Object
329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 |
# File 'lib/dead_bro/sql_subscriber.rb', line 329 def self.interpolate_sql_with_binds(sql, binds, connection) return sql if binds.nil? || binds.empty? return sql unless connection interpolated_sql = sql.dup # Handle $1, $2 style placeholders (PostgreSQL) if interpolated_sql.include?("$1") binds.each_with_index do |val, index| # Get value from bind param (handle both raw values and ActiveRecord::Relation::QueryAttribute) value = val if val.respond_to?(:value_for_database) value = val.value_for_database elsif val.respond_to?(:value) value = val.value end quoted_value = connection.quote(value) interpolated_sql = interpolated_sql.gsub("$#{index + 1}", quoted_value) end elsif interpolated_sql.include?("?") # Handle ? style placeholders (MySQL, SQLite) # We need to replace ? one by one in order binds.each do |val| # Get value from bind param value = val if val.respond_to?(:value_for_database) value = val.value_for_database elsif val.respond_to?(:value) value = val.value end quoted_value = connection.quote(value) interpolated_sql = interpolated_sql.sub("?", quoted_value) end end interpolated_sql end |
.safe_query_trace(data, captured_backtrace = nil) ⇒ Object
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 |
# File 'lib/dead_bro/sql_subscriber.rb', line 460 def self.safe_query_trace(data, captured_backtrace = nil) return [] unless data.is_a?(Hash) # Build trace from available data fields trace = [] # Use filename, line, and method if available if data[:filename] && data[:line] && data[:method] trace << "#{data[:filename]}:#{data[:line]}:in `#{data[:method]}'" end # Use the captured backtrace from when the query started (most accurate) if captured_backtrace && captured_backtrace.is_a?(Array) && !captured_backtrace.empty? # Filter to only include frames that contain "app/" (application code) app_frames = captured_backtrace.select do |frame| frame.include?("app/") && !frame.include?("/vendor/") end caller_trace = app_frames.map do |line| # Remove any potential sensitive information from file paths line.gsub(/\/[^\/]*(password|secret|key|token)[^\/]*\//i, "/[FILTERED]/") end trace.concat(caller_trace) else # Fallback: try to get backtrace from current context begin # Get all available frames - we'll filter to find application code all_frames = Thread.current.backtrace || [] if all_frames.empty? # Fallback to caller_locations if backtrace is empty locations = caller_locations(1, 50) all_frames = locations.map { |loc| "#{loc.path}:#{loc.lineno}:in `#{loc.label}'" } if locations end # Filter to only include frames that contain "app/" (application code) app_frames = all_frames.select do |frame| frame.include?("app/") && !frame.include?("/vendor/") end caller_trace = app_frames.map do |line| line.gsub(/\/[^\/]*(password|secret|key|token)[^\/]*\//i, "/[FILTERED]/") end trace.concat(caller_trace) rescue # If backtrace fails, try caller as fallback begin caller_stack = caller(20, 50) # Get more frames to find app/ frames app_frames = caller_stack.select { |frame| frame.include?("app/") && !frame.include?("/vendor/") } caller_trace = app_frames.map do |line| line.gsub(/\/[^\/]*(password|secret|key|token)[^\/]*\//i, "/[FILTERED]/") end trace.concat(caller_trace) rescue # If caller also fails, we still have the immediate location end end end # If we have a backtrace in the data, use it (but it's usually nil for SQL events) if data[:backtrace] && data[:backtrace].is_a?(Array) # Filter to only include frames that contain "app/" app_backtrace = data[:backtrace].select do |line| line.is_a?(String) && line.include?("app/") && !line.include?("/vendor/") end backtrace_trace = app_backtrace.map do |line| case line when String line.gsub(/\/[^\/]*(password|secret|key|token)[^\/]*\//i, "/[FILTERED]/") else line.to_s end end trace.concat(backtrace_trace) end # Remove duplicates and return all app/ frames (no limit) trace.uniq.map do |line| case line when String # Remove any potential sensitive information from file paths line.gsub(/\/[^\/]*(password|secret|key|token)[^\/]*\//i, "/[FILTERED]/") else line.to_s end end rescue [] end |
.sanitize_sql(sql) ⇒ Object
175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 |
# File 'lib/dead_bro/sql_subscriber.rb', line 175 def self.sanitize_sql(sql) return sql unless sql.is_a?(String) # Cap length first — most "expensive" queries from the app's perspective # are big UPDATE/INSERT with long literal blobs; don't burn regex time on # those when we're going to truncate anyway. sql = sql[0..SANITIZE_MAX_LENGTH] + "..." if sql.length > SANITIZE_MAX_LENGTH # Only scan for sensitive KV pairs if one of the keywords is actually # present — saves two regex passes on the vast majority of queries. if sql.match?(SANITIZE_SKIP_SENSITIVE_WHEN_NO_KEYWORDS) sql = sql.gsub(SENSITIVE_KV_QUOTED_RE, '\1 = ?') sql = sql.gsub(SENSITIVE_KV_BARE_RE, '\1 = ?') end # Same short-circuit for WHERE rewrite. if sql.match?(SANITIZE_SKIP_WHERE_WHEN_NO_KEYWORD) sql = sql.gsub(WHERE_EQ_QUOTED_RE) do |match| match.gsub(WHERE_EQ_QUOTED_INNER_RE, "= ?") end end sql end |
.should_continue_tracking?(current_queries_array, max_count) ⇒ Boolean
Check if we should continue tracking based on count and time limits
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 |
# File 'lib/dead_bro/sql_subscriber.rb', line 44 def self.should_continue_tracking?(current_queries_array, max_count) return false unless current_queries_array.is_a?(Array) # Check count limit return false if current_queries_array.length >= max_count # Check time limit start_time = Thread.current[DeadBro::TRACKING_START_TIME_KEY] if start_time elapsed_seconds = Time.now - start_time return false if elapsed_seconds >= DeadBro::MAX_TRACKING_DURATION_SECONDS end true end |
.should_explain_query?(duration_ms, sql) ⇒ Boolean
200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 |
# File 'lib/dead_bro/sql_subscriber.rb', line 200 def self.should_explain_query?(duration_ms, sql) return false unless DeadBro.configuration.explain_analyze_enabled return false if duration_ms < DeadBro.configuration.slow_query_threshold_ms return false unless sql.is_a?(String) return false if sql.strip.empty? # Skip EXPLAIN for certain query types that don't benefit from it sql_upper = sql.upcase.strip return false if sql_upper.start_with?("EXPLAIN") return false if sql_upper.start_with?("BEGIN") return false if sql_upper.start_with?("COMMIT") return false if sql_upper.start_with?("ROLLBACK") return false if sql_upper.start_with?("SAVEPOINT") return false if sql_upper.start_with?("RELEASE") true end |
.start_explain_analyze_background(sql, connection_id, query_info, binds = nil) ⇒ Object
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 |
# File 'lib/dead_bro/sql_subscriber.rb', line 218 def self.start_explain_analyze_background(sql, connection_id, query_info, binds = nil) return unless defined?(ActiveRecord) return unless ActiveRecord::Base.respond_to?(:connection) # Cap pending EXPLAINs per request. A slow-query storm that would have # spawned 200 threads and starved the AR pool now drops excess plans # instead of cascading into a timeout. pending = Thread.current[THREAD_LOCAL_EXPLAIN_PENDING_KEY] ||= [] if pending.length >= MAX_PENDING_EXPLAINS query_info[:explain_plan] = nil return end # Capture the main thread reference to append logs to the correct thread main_thread = Thread.current # Run EXPLAIN in a background thread to avoid blocking the main request. # We use `with_connection` so the connection returns to the pool even if # the thread is killed or the block raises — the previous manual # checkout/checkin could leak a connection under pathological paths. explain_thread = Thread.new do begin ActiveRecord::Base.connection_pool.with_connection do |connection| final_sql = interpolate_sql_with_binds(sql, binds, connection) explain_sql = build_explain_query(final_sql, connection) adapter_name = connection.adapter_name.downcase result = if adapter_name == "postgresql" || adapter_name == "postgis" connection.select_all(explain_sql) else connection.execute(explain_sql) end explain_plan = format_explain_result(result, connection) query_info[:explain_plan] = if explain_plan && !explain_plan.to_s.strip.empty? explain_plan end end rescue => e # Silently fail — don't let EXPLAIN break the application. append_log_to_thread(main_thread, :debug, "Failed to capture EXPLAIN ANALYZE: #{e.}") query_info[:explain_plan] = nil end end pending << explain_thread rescue => e # Use DeadBro.logger here since we're still in the main thread DeadBro.logger.debug("Failed to start EXPLAIN ANALYZE thread: #{e.}") end |
.start_request_tracking ⇒ Object
122 123 124 125 126 127 128 129 |
# File 'lib/dead_bro/sql_subscriber.rb', line 122 def self.start_request_tracking # Stack allows nested job tracking (e.g. one job performing others in the same thread) (Thread.current[THREAD_LOCAL_KEY] ||= []) << [] Thread.current[THREAD_LOCAL_ALLOC_START_KEY] = {} Thread.current[THREAD_LOCAL_ALLOC_RESULTS_KEY] = {} Thread.current[THREAD_LOCAL_BACKTRACE_KEY] = {} Thread.current[THREAD_LOCAL_EXPLAIN_PENDING_KEY] = [] end |
.stop_request_tracking ⇒ Object
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 |
# File 'lib/dead_bro/sql_subscriber.rb', line 131 def self.stop_request_tracking # Wait for any pending EXPLAIN ANALYZE queries to complete (with timeout) # This must happen BEFORE we get the queries array reference to ensure # all explain_plan fields are populated wait_for_pending_explains(EXPLAIN_WAIT_TIMEOUT_SECONDS) stack = Thread.current[THREAD_LOCAL_KEY] queries = (stack.is_a?(Array) && stack.any?) ? stack.pop : [] # Clear thread locals when stack is empty so "tracking not started" behaves correctly if stack.nil? || stack.empty? Thread.current[THREAD_LOCAL_KEY] = nil Thread.current[THREAD_LOCAL_ALLOC_START_KEY] = nil Thread.current[THREAD_LOCAL_ALLOC_RESULTS_KEY] = nil Thread.current[THREAD_LOCAL_BACKTRACE_KEY] = nil Thread.current[THREAD_LOCAL_EXPLAIN_PENDING_KEY] = nil end queries end |
.subscribe! ⇒ Object
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 |
# File 'lib/dead_bro/sql_subscriber.rb', line 60 def self.subscribe! # Subscribe with a start/finish listener to measure allocations per query if ActiveSupport::Notifications.notifier.respond_to?(:subscribe) begin ActiveSupport::Notifications.notifier.subscribe(SQL_EVENT_NAME, SqlAllocListener.new) rescue end end ActiveSupport::Notifications.subscribe(SQL_EVENT_NAME) do |name, started, finished, _unique_id, data| next if data[:name] == "SCHEMA" || data[:name] == "CACHE" || data[:name] == "BEGIN" || data[:name] == "COMMIT" || data[:name] == "ROLLBACK" || data[:name] == "SAVEPOINT" || data[:name] == "RELEASE" # Only track queries that are part of the current request (top of stack for nested jobs) current = current_queries_array next unless current unique_id = _unique_id allocations = nil begin alloc_results = Thread.current[THREAD_LOCAL_ALLOC_RESULTS_KEY] allocations = alloc_results && alloc_results.delete(unique_id) rescue end duration_ms = ((finished - started) * 1000.0).round(2) original_sql = data[:sql] # Only capture a backtrace for queries we actually care about tracing # (slow). This skips the ~O(stack-depth) allocation on the 99% of queries # that are fast. An N+1 of 100 x 1ms queries no longer eats a thousand # frame allocations for traces nobody will read. threshold = begin DeadBro.configuration.slow_query_threshold_ms rescue 500 end captured_trace = (duration_ms >= threshold.to_f) ? capture_app_backtrace : [] query_info = { sql: sanitize_sql(original_sql), name: data[:name], duration_ms: duration_ms, cached: data[:cached] || false, connection_id: data[:connection_id], trace: captured_trace, allocations: allocations } # Run EXPLAIN ANALYZE for slow queries in the background if should_explain_query?(duration_ms, original_sql) # Store reference to query_info so we can update it when EXPLAIN completes query_info[:explain_plan] = nil # Placeholder # Capture binds if available (type_casted_binds is preferred as they are ready for quoting) binds = data[:type_casted_binds] || data[:binds] start_explain_analyze_background(original_sql, data[:connection_id], query_info, binds) end # Add to current context (top of stack), but only if we haven't exceeded the limits if should_continue_tracking?(current, MAX_TRACKED_QUERIES) current << query_info end end end |
.tracking_active? ⇒ Boolean
True when there is at least one active tracking context (e.g. for nested jobs).
31 32 33 34 |
# File 'lib/dead_bro/sql_subscriber.rb', line 31 def self.tracking_active? stack = Thread.current[THREAD_LOCAL_KEY] stack.is_a?(Array) && stack.any? end |
.wait_for_pending_explains(timeout_seconds) ⇒ Object
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 |
# File 'lib/dead_bro/sql_subscriber.rb', line 158 def self.wait_for_pending_explains(timeout_seconds) pending = Thread.current[THREAD_LOCAL_EXPLAIN_PENDING_KEY] return unless pending && !pending.empty? start_time = Process.clock_gettime(Process::CLOCK_MONOTONIC) pending.each do |thread| remaining_time = timeout_seconds - (Process.clock_gettime(Process::CLOCK_MONOTONIC) - start_time) break if remaining_time <= 0 begin thread.join(remaining_time) rescue => e DeadBro.logger.debug("Error waiting for EXPLAIN ANALYZE: #{e.}") end end end |