| 1 | #!/usr/bin/env bash |
| 2 | set -euo pipefail |
| 3 | |
| 4 | SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" |
| 5 | # shellcheck disable=SC1091 |
| 6 | . "$SCRIPT_DIR/lib.sh" |
| 7 | |
| 8 | init_workspace "mssql" |
| 9 | trap cleanup EXIT |
| 10 | |
| 11 | MSSQL_PORT="$(reserve_port)" |
| 12 | write_env "MSSQL_PORT" "$MSSQL_PORT" |
| 13 | MSSQL_CONF="$WORKDIR/config/go.d/mssql.conf" |
| 14 | replace_in_file "$MSSQL_CONF" "127.0.0.1:1433" "127.0.0.1:${MSSQL_PORT}" |
| 15 | |
| 16 | MSSQL_VARIANT_LABEL="${MSSQL_VARIANT:-mssql}" |
| 17 | if [ -n "${MSSQL_IMAGE:-}" ]; then |
| 18 | write_env "MSSQL_IMAGE" "$MSSQL_IMAGE" |
| 19 | fi |
| 20 | |
| 21 | compose_up mssql |
| 22 | wait_healthy mssql 120 |
| 23 | compose_run mssql-init |
| 24 | |
| 25 | build_plugin |
| 26 | |
| 27 | MSSQL_JOB_RETRIES="${MSSQL_JOB_RETRIES:-6}" |
| 28 | MSSQL_JOB_RETRY_DELAY="${MSSQL_JOB_RETRY_DELAY:-5}" |
| 29 | |
| 30 | mssql_is_no_jobs_started() { |
| 31 | local input="$1" |
| 32 | if [ ! -s "$input" ]; then |
| 33 | return 1 |
| 34 | fi |
| 35 | |
| 36 | if command -v python3 >/dev/null 2>&1; then |
| 37 | python3 - "$input" <<'PY' |
| 38 | import json |
| 39 | import sys |
| 40 | |
| 41 | path = sys.argv[1] |
| 42 | try: |
| 43 | with open(path, "r", encoding="utf-8") as fh: |
| 44 | doc = json.load(fh) |
| 45 | except Exception: |
| 46 | raise SystemExit(1) |
| 47 | |
| 48 | status = doc.get("status") |
| 49 | msg = str(doc.get("errorMessage") or "").lower() |
| 50 | if status == 503 and "no jobs started for module" in msg: |
| 51 | raise SystemExit(0) |
| 52 | raise SystemExit(1) |
| 53 | PY |
| 54 | return $? |
| 55 | fi |
| 56 | |
| 57 | python - "$input" <<'PY' |
| 58 | import json |
| 59 | import sys |
| 60 | |
| 61 | path = sys.argv[1] |
| 62 | try: |
| 63 | with open(path, "r") as fh: |
| 64 | doc = json.load(fh) |
| 65 | except Exception: |
| 66 | raise SystemExit(1) |
| 67 | |
| 68 | status = doc.get("status") |
| 69 | msg = str(doc.get("errorMessage") or "").lower() |
| 70 | if status == 503 and "no jobs started for module" in msg: |
| 71 | raise SystemExit(0) |
| 72 | raise SystemExit(1) |
| 73 | PY |
| 74 | } |
| 75 | |
| 76 | run_mssql_info_with_retry() { |
| 77 | local output="$WORKDIR/mssql-top-queries-info.json" |
| 78 | local attempt=1 |
| 79 | |
| 80 | while true; do |
| 81 | if run "$WORKDIR/go.d.plugin" \ |
| 82 | --config-dir "$WORKDIR/config" \ |
| 83 | --function "mssql:top-queries" \ |
| 84 | --function-args info \ |
| 85 | > "$output"; then |
| 86 | validate "$output" |
| 87 | return 0 |
| 88 | fi |
| 89 | |
| 90 | if mssql_is_no_jobs_started "$output"; then |
| 91 | if [ "$attempt" -ge "$MSSQL_JOB_RETRIES" ]; then |
| 92 | echo "Timed out waiting for mssql jobs to start (info)" >&2 |
| 93 | return 1 |
| 94 | fi |
| 95 | attempt=$((attempt + 1)) |
| 96 | sleep "$MSSQL_JOB_RETRY_DELAY" |
| 97 | continue |
| 98 | fi |
| 99 | |
| 100 | echo "Unexpected failure while running mssql top-queries info" >&2 |
| 101 | cat "$output" >&2 |
| 102 | return 1 |
| 103 | done |
| 104 | } |
| 105 | |
| 106 | run_mssql_top_queries_with_retry() { |
| 107 | local output="$WORKDIR/mssql-top-queries.json" |
| 108 | local attempt=1 |
| 109 | |
| 110 | while true; do |
| 111 | if run "$WORKDIR/go.d.plugin" \ |
| 112 | --config-dir "$WORKDIR/config" \ |
| 113 | --function "mssql:top-queries" \ |
| 114 | --function-args __job:local \ |
| 115 | > "$output"; then |
| 116 | validate "$output" --min-rows 1 |
| 117 | return 0 |
| 118 | fi |
| 119 | |
| 120 | if mssql_is_no_jobs_started "$output"; then |
| 121 | if [ "$attempt" -ge "$MSSQL_JOB_RETRIES" ]; then |
| 122 | echo "Timed out waiting for mssql jobs to start (top-queries)" >&2 |
| 123 | return 1 |
| 124 | fi |
| 125 | attempt=$((attempt + 1)) |
| 126 | sleep "$MSSQL_JOB_RETRY_DELAY" |
| 127 | continue |
| 128 | fi |
| 129 | |
| 130 | echo "Unexpected failure while running mssql top-queries" >&2 |
| 131 | cat "$output" >&2 |
| 132 | return 1 |
| 133 | done |
| 134 | } |
| 135 | |
| 136 | run_mssql_function_with_retry() { |
| 137 | local method="$1" |
| 138 | local args="${2:-__job:local}" |
| 139 | local require_rows="${3:-true}" |
| 140 | local output="$WORKDIR/mssql-${method}.json" |
| 141 | local attempt=1 |
| 142 | |
| 143 | while true; do |
| 144 | if run "$WORKDIR/go.d.plugin" \ |
| 145 | --config-dir "$WORKDIR/config" \ |
| 146 | --function "mssql:${method}" \ |
| 147 | --function-args "$args" \ |
| 148 | > "$output"; then |
| 149 | if [ "$require_rows" = "true" ]; then |
| 150 | validate "$output" --min-rows 1 |
| 151 | else |
| 152 | validate "$output" |
| 153 | fi |
| 154 | echo "$output" |
| 155 | return 0 |
| 156 | fi |
| 157 | |
| 158 | if mssql_is_no_jobs_started "$output"; then |
| 159 | if [ "$attempt" -ge "$MSSQL_JOB_RETRIES" ]; then |
| 160 | echo "Timed out waiting for mssql jobs to start (${method})" >&2 |
| 161 | return 1 |
| 162 | fi |
| 163 | attempt=$((attempt + 1)) |
| 164 | sleep "$MSSQL_JOB_RETRY_DELAY" |
| 165 | continue |
| 166 | fi |
| 167 | |
| 168 | echo "Unexpected failure while running mssql ${method}" >&2 |
| 169 | cat "$output" >&2 |
| 170 | return 1 |
| 171 | done |
| 172 | } |
| 173 | |
| 174 | run_mssql_info_with_retry |
| 175 | run_mssql_top_queries_with_retry |
| 176 | |
| 177 | mssql_container_id() { |
| 178 | "${COMPOSE[@]}" ps -q mssql |
| 179 | } |
| 180 | |
| 181 | mssql_sqlcmd_path() { |
| 182 | local cid |
| 183 | cid="$(mssql_container_id)" |
| 184 | if [ -z "$cid" ]; then |
| 185 | echo "MSSQL container ID not found" >&2 |
| 186 | return 1 |
| 187 | fi |
| 188 | docker exec -i "$cid" bash -lc 'if [ -x /opt/mssql-tools18/bin/sqlcmd ]; then echo /opt/mssql-tools18/bin/sqlcmd; elif [ -x /opt/mssql-tools/bin/sqlcmd ]; then echo /opt/mssql-tools/bin/sqlcmd; else exit 1; fi' |
| 189 | } |
| 190 | |
| 191 | MSSQL_SQLCMD_PATH="$(mssql_sqlcmd_path)" |
| 192 | echo "Using sqlcmd path: $MSSQL_SQLCMD_PATH" >&2 |
| 193 | |
| 194 | mssql_sqlcmd_supports_c() { |
| 195 | local cid |
| 196 | cid="$(mssql_container_id)" |
| 197 | if [ -z "$cid" ]; then |
| 198 | return 1 |
| 199 | fi |
| 200 | set +e |
| 201 | local help |
| 202 | help="$(docker exec -i "$cid" "$MSSQL_SQLCMD_PATH" -? 2>&1)" |
| 203 | local status=$? |
| 204 | set -e |
| 205 | if [ $status -ne 0 ] && [ -z "$help" ]; then |
| 206 | return 1 |
| 207 | fi |
| 208 | echo "$help" | grep -q " -C" |
| 209 | } |
| 210 | |
| 211 | MSSQL_SQLCMD_CFLAG=() |
| 212 | case "$MSSQL_SQLCMD_PATH" in |
| 213 | *mssql-tools18*) |
| 214 | MSSQL_SQLCMD_CFLAG=(-C) |
| 215 | ;; |
| 216 | *) |
| 217 | if mssql_sqlcmd_supports_c; then |
| 218 | MSSQL_SQLCMD_CFLAG=(-C) |
| 219 | fi |
| 220 | ;; |
| 221 | esac |
| 222 | |
| 223 | mssql_exec_sa() { |
| 224 | local sql="$1" |
| 225 | local cid |
| 226 | cid="$(mssql_container_id)" |
| 227 | if [ -z "$cid" ]; then |
| 228 | echo "MSSQL container ID not found" >&2 |
| 229 | return 1 |
| 230 | fi |
| 231 | run docker exec -i "$cid" "$MSSQL_SQLCMD_PATH" "${MSSQL_SQLCMD_CFLAG[@]}" -S localhost -U sa -P "Netdata123!" -d netdata -b -y 0 -Y 0 -Q "$sql" |
| 232 | } |
| 233 | |
| 234 | mssql_exec_sa_allow_error() { |
| 235 | local sql="$1" |
| 236 | local cid |
| 237 | cid="$(mssql_container_id)" |
| 238 | if [ -z "$cid" ]; then |
| 239 | echo "MSSQL container ID not found" >&2 |
| 240 | return 1 |
| 241 | fi |
| 242 | set +e |
| 243 | docker exec -i "$cid" "$MSSQL_SQLCMD_PATH" "${MSSQL_SQLCMD_CFLAG[@]}" -S localhost -U sa -P "Netdata123!" -d netdata -b -y 0 -Y 0 -Q "$sql" >/dev/null 2>&1 |
| 244 | set -e |
| 245 | } |
| 246 | |
| 247 | induce_deadlock_once() { |
| 248 | local tx1 |
| 249 | local tx2 |
| 250 | |
| 251 | tx1="$(cat <<'SQL' |
| 252 | SET NOCOUNT ON; |
| 253 | SET LOCK_TIMEOUT 5000; |
| 254 | BEGIN TRAN; |
| 255 | UPDATE dbo.deadlock_a SET value = value + 1 WHERE id = 1; |
| 256 | WAITFOR DELAY '00:00:01'; |
| 257 | UPDATE dbo.deadlock_b SET value = value + 1 WHERE id = 1; |
| 258 | COMMIT; |
| 259 | SQL |
| 260 | )" |
| 261 | |
| 262 | tx2="$(cat <<'SQL' |
| 263 | SET NOCOUNT ON; |
| 264 | SET LOCK_TIMEOUT 5000; |
| 265 | BEGIN TRAN; |
| 266 | UPDATE dbo.deadlock_b SET value = value + 1 WHERE id = 1; |
| 267 | WAITFOR DELAY '00:00:01'; |
| 268 | UPDATE dbo.deadlock_a SET value = value + 1 WHERE id = 1; |
| 269 | COMMIT; |
| 270 | SQL |
| 271 | )" |
| 272 | |
| 273 | mssql_exec_sa "$tx1" & |
| 274 | local pid1=$! |
| 275 | mssql_exec_sa "$tx2" & |
| 276 | local pid2=$! |
| 277 | |
| 278 | set +e |
| 279 | wait "$pid1" |
| 280 | wait "$pid2" |
| 281 | set -e |
| 282 | } |
| 283 | |
| 284 | assert_deadlock_info_content() { |
| 285 | local input="$1" |
| 286 | if command -v python3 >/dev/null 2>&1; then |
| 287 | python3 - "$input" <<'PY' |
| 288 | import json |
| 289 | import re |
| 290 | import sys |
| 291 | |
| 292 | path = sys.argv[1] |
| 293 | with open(path, "r", encoding="utf-8") as fh: |
| 294 | doc = json.load(fh) |
| 295 | |
| 296 | try: |
| 297 | status = int(doc.get("status")) |
| 298 | except (TypeError, ValueError): |
| 299 | raise SystemExit(f"unexpected status value: {doc.get('status')!r}") |
| 300 | |
| 301 | if status != 200: |
| 302 | raise SystemExit(f"expected status 200, got {status}") |
| 303 | |
| 304 | if doc.get("errorMessage"): |
| 305 | raise SystemExit(f"unexpected errorMessage on status 200: {doc.get('errorMessage')!r}") |
| 306 | |
| 307 | columns = doc.get("columns") or {} |
| 308 | field_to_idx = {} |
| 309 | if isinstance(columns, dict): |
| 310 | for field, col in columns.items(): |
| 311 | if not isinstance(col, dict): |
| 312 | continue |
| 313 | try: |
| 314 | field_to_idx[field] = int(col.get("index")) |
| 315 | except (TypeError, ValueError): |
| 316 | continue |
| 317 | else: |
| 318 | for idx, col in enumerate(columns): |
| 319 | if not isinstance(col, dict): |
| 320 | continue |
| 321 | field = col.get("field") |
| 322 | if field: |
| 323 | field_to_idx[field] = idx |
| 324 | |
| 325 | for required in ("row_id", "deadlock_id", "process_id", "is_victim", "lock_mode", "lock_status", "query_text", "wait_resource", "database"): |
| 326 | if required not in field_to_idx: |
| 327 | raise SystemExit(f"missing expected column: {required}") |
| 328 | |
| 329 | data = doc.get("data") or [] |
| 330 | if not data: |
| 331 | raise SystemExit("deadlock-info returned no rows") |
| 332 | |
| 333 | def get_value(row, field): |
| 334 | idx = field_to_idx[field] |
| 335 | return row[idx] if idx < len(row) else None |
| 336 | |
| 337 | def norm(val): |
| 338 | return "" if val is None else str(val).strip() |
| 339 | |
| 340 | has_waiting = any(str(get_value(row, "lock_status")).upper() == "WAITING" for row in data) |
| 341 | if not has_waiting: |
| 342 | raise SystemExit("no WAITING lock_status found in deadlock-info output") |
| 343 | |
| 344 | table_pattern = re.compile(r"deadlock_(a|b)", re.IGNORECASE) |
| 345 | has_expected_query = any(table_pattern.search(str(get_value(row, "query_text"))) for row in data) |
| 346 | if not has_expected_query: |
| 347 | raise SystemExit("query_text does not reference deadlock tables") |
| 348 | |
| 349 | waiting_rows = [row for row in data if str(get_value(row, "lock_status")).upper() == "WAITING"] |
| 350 | if any(norm(get_value(row, "lock_mode")) == "" for row in waiting_rows): |
| 351 | raise SystemExit("WAITING rows must include lock_mode") |
| 352 | if any(norm(get_value(row, "wait_resource")) == "" for row in waiting_rows): |
| 353 | raise SystemExit("WAITING rows must include wait_resource") |
| 354 | |
| 355 | lock_mode_re = re.compile(r"^[A-Za-z0-9_-]+$") |
| 356 | if any(not lock_mode_re.match(norm(get_value(row, "lock_mode"))) for row in waiting_rows): |
| 357 | raise SystemExit("WAITING rows must include a valid lock_mode") |
| 358 | |
| 359 | victim_counts = {} |
| 360 | expected_db = "netdata" |
| 361 | has_database = False |
| 362 | for row in data: |
| 363 | deadlock_id = norm(get_value(row, "deadlock_id")) |
| 364 | if deadlock_id == "": |
| 365 | raise SystemExit("deadlock_id missing from deadlock-info output") |
| 366 | process_id = norm(get_value(row, "process_id")) |
| 367 | if process_id == "": |
| 368 | raise SystemExit("process_id missing from deadlock-info output") |
| 369 | row_id = norm(get_value(row, "row_id")) |
| 370 | if row_id != f"{deadlock_id}:{process_id}": |
| 371 | raise SystemExit(f"row_id {row_id} does not match deadlock_id/process_id") |
| 372 | victim_counts.setdefault(deadlock_id, 0) |
| 373 | if str(get_value(row, "is_victim")).lower() == "true": |
| 374 | victim_counts[deadlock_id] += 1 |
| 375 | db_val = norm(get_value(row, "database")).lower() |
| 376 | if db_val: |
| 377 | has_database = True |
| 378 | if db_val != expected_db: |
| 379 | raise SystemExit(f"unexpected database value {db_val!r}, expected {expected_db!r}") |
| 380 | |
| 381 | for deadlock_id, count in victim_counts.items(): |
| 382 | if count != 1: |
| 383 | raise SystemExit(f"deadlock_id {deadlock_id} has victim count {count}, expected 1") |
| 384 | if not has_database: |
| 385 | raise SystemExit("expected at least one row with database populated") |
| 386 | PY |
| 387 | return |
| 388 | fi |
| 389 | |
| 390 | python - "$input" <<'PY' |
| 391 | import json |
| 392 | import re |
| 393 | import sys |
| 394 | |
| 395 | path = sys.argv[1] |
| 396 | with open(path, "r") as fh: |
| 397 | doc = json.load(fh) |
| 398 | |
| 399 | try: |
| 400 | status = int(doc.get("status")) |
| 401 | except (TypeError, ValueError): |
| 402 | raise SystemExit("unexpected status value: %r" % (doc.get("status"),)) |
| 403 | |
| 404 | if status != 200: |
| 405 | raise SystemExit("expected status 200, got %s" % status) |
| 406 | |
| 407 | if doc.get("errorMessage"): |
| 408 | raise SystemExit("unexpected errorMessage on status 200: %r" % (doc.get("errorMessage"),)) |
| 409 | |
| 410 | columns = doc.get("columns") or {} |
| 411 | field_to_idx = {} |
| 412 | if isinstance(columns, dict): |
| 413 | for field, col in columns.items(): |
| 414 | if not isinstance(col, dict): |
| 415 | continue |
| 416 | try: |
| 417 | field_to_idx[field] = int(col.get("index")) |
| 418 | except (TypeError, ValueError): |
| 419 | continue |
| 420 | else: |
| 421 | for idx, col in enumerate(columns): |
| 422 | if not isinstance(col, dict): |
| 423 | continue |
| 424 | field = col.get("field") |
| 425 | if field: |
| 426 | field_to_idx[field] = idx |
| 427 | |
| 428 | for required in ("row_id", "deadlock_id", "process_id", "is_victim", "lock_mode", "lock_status", "query_text", "wait_resource", "database"): |
| 429 | if required not in field_to_idx: |
| 430 | raise SystemExit("missing expected column: %s" % required) |
| 431 | |
| 432 | data = doc.get("data") or [] |
| 433 | if not data: |
| 434 | raise SystemExit("deadlock-info returned no rows") |
| 435 | |
| 436 | def get_value(row, field): |
| 437 | idx = field_to_idx[field] |
| 438 | return row[idx] if idx < len(row) else None |
| 439 | |
| 440 | def norm(val): |
| 441 | return "" if val is None else str(val).strip() |
| 442 | |
| 443 | has_waiting = any(str(get_value(row, "lock_status")).upper() == "WAITING" for row in data) |
| 444 | if not has_waiting: |
| 445 | raise SystemExit("no WAITING lock_status found in deadlock-info output") |
| 446 | |
| 447 | table_pattern = re.compile(r"deadlock_(a|b)", re.IGNORECASE) |
| 448 | has_expected_query = any(table_pattern.search(str(get_value(row, "query_text"))) for row in data) |
| 449 | if not has_expected_query: |
| 450 | raise SystemExit("query_text does not reference deadlock tables") |
| 451 | |
| 452 | waiting_rows = [row for row in data if str(get_value(row, "lock_status")).upper() == "WAITING"] |
| 453 | if any(norm(get_value(row, "lock_mode")) == "" for row in waiting_rows): |
| 454 | raise SystemExit("WAITING rows must include lock_mode") |
| 455 | if any(norm(get_value(row, "wait_resource")) == "" for row in waiting_rows): |
| 456 | raise SystemExit("WAITING rows must include wait_resource") |
| 457 | |
| 458 | lock_mode_re = re.compile(r"^[A-Za-z0-9_-]+$") |
| 459 | if any(not lock_mode_re.match(norm(get_value(row, "lock_mode"))) for row in waiting_rows): |
| 460 | raise SystemExit("WAITING rows must include a valid lock_mode") |
| 461 | |
| 462 | victim_counts = {} |
| 463 | expected_db = "netdata" |
| 464 | has_database = False |
| 465 | for row in data: |
| 466 | deadlock_id = norm(get_value(row, "deadlock_id")) |
| 467 | if deadlock_id == "": |
| 468 | raise SystemExit("deadlock_id missing from deadlock-info output") |
| 469 | process_id = norm(get_value(row, "process_id")) |
| 470 | if process_id == "": |
| 471 | raise SystemExit("process_id missing from deadlock-info output") |
| 472 | row_id = norm(get_value(row, "row_id")) |
| 473 | if row_id != "%s:%s" % (deadlock_id, process_id): |
| 474 | raise SystemExit("row_id %s does not match deadlock_id/process_id" % row_id) |
| 475 | victim_counts.setdefault(deadlock_id, 0) |
| 476 | if str(get_value(row, "is_victim")).lower() == "true": |
| 477 | victim_counts[deadlock_id] += 1 |
| 478 | db_val = norm(get_value(row, "database")).lower() |
| 479 | if db_val: |
| 480 | has_database = True |
| 481 | if db_val != expected_db: |
| 482 | raise SystemExit("unexpected database value %r, expected %r" % (db_val, expected_db)) |
| 483 | |
| 484 | for deadlock_id, count in victim_counts.items(): |
| 485 | if count != 1: |
| 486 | raise SystemExit("deadlock_id %s has victim count %s, expected 1" % (deadlock_id, count)) |
| 487 | if not has_database: |
| 488 | raise SystemExit("expected at least one row with database populated") |
| 489 | PY |
| 490 | } |
| 491 | |
| 492 | assert_deadlock_info_empty_success() { |
| 493 | local input="$1" |
| 494 | |
| 495 | if command -v python3 >/dev/null 2>&1; then |
| 496 | python3 - "$input" <<'PY' |
| 497 | import json |
| 498 | import sys |
| 499 | |
| 500 | path = sys.argv[1] |
| 501 | with open(path, "r", encoding="utf-8") as fh: |
| 502 | doc = json.load(fh) |
| 503 | |
| 504 | try: |
| 505 | status = int(doc.get("status")) |
| 506 | except (TypeError, ValueError): |
| 507 | raise SystemExit(f"unexpected status value: {doc.get('status')!r}") |
| 508 | |
| 509 | if status != 200: |
| 510 | raise SystemExit(f"expected status 200, got {status}") |
| 511 | |
| 512 | if doc.get("errorMessage"): |
| 513 | raise SystemExit(f"unexpected errorMessage on status 200: {doc.get('errorMessage')!r}") |
| 514 | |
| 515 | columns = doc.get("columns") or {} |
| 516 | field_to_idx = {} |
| 517 | if isinstance(columns, dict): |
| 518 | for field, col in columns.items(): |
| 519 | if not isinstance(col, dict): |
| 520 | continue |
| 521 | try: |
| 522 | field_to_idx[field] = int(col.get("index")) |
| 523 | except (TypeError, ValueError): |
| 524 | continue |
| 525 | else: |
| 526 | for idx, col in enumerate(columns): |
| 527 | if not isinstance(col, dict): |
| 528 | continue |
| 529 | field = col.get("field") |
| 530 | if field: |
| 531 | field_to_idx[field] = idx |
| 532 | |
| 533 | data = doc.get("data") or [] |
| 534 | if len(data) == 0: |
| 535 | raise SystemExit(0) |
| 536 | |
| 537 | query_idx = field_to_idx.get("query_text", None) |
| 538 | if query_idx is None: |
| 539 | raise SystemExit(f"expected no rows, got {len(data)}") |
| 540 | |
| 541 | for row in data: |
| 542 | if query_idx >= len(row): |
| 543 | continue |
| 544 | query = str(row[query_idx]).lower() |
| 545 | if "deadlock_a" in query or "deadlock_b" in query: |
| 546 | raise SystemExit(f"unexpected deadlock rows for test tables, got {len(data)} rows") |
| 547 | PY |
| 548 | return |
| 549 | fi |
| 550 | |
| 551 | python - "$input" <<'PY' |
| 552 | import json |
| 553 | import sys |
| 554 | |
| 555 | path = sys.argv[1] |
| 556 | with open(path, "r") as fh: |
| 557 | doc = json.load(fh) |
| 558 | |
| 559 | try: |
| 560 | status = int(doc.get("status")) |
| 561 | except (TypeError, ValueError): |
| 562 | raise SystemExit("unexpected status value: %r" % (doc.get("status"),)) |
| 563 | |
| 564 | if status != 200: |
| 565 | raise SystemExit("expected status 200, got %s" % status) |
| 566 | |
| 567 | if doc.get("errorMessage"): |
| 568 | raise SystemExit("unexpected errorMessage on status 200: %r" % (doc.get("errorMessage"),)) |
| 569 | |
| 570 | columns = doc.get("columns") or {} |
| 571 | field_to_idx = {} |
| 572 | if isinstance(columns, dict): |
| 573 | for field, col in columns.items(): |
| 574 | if not isinstance(col, dict): |
| 575 | continue |
| 576 | try: |
| 577 | field_to_idx[field] = int(col.get("index")) |
| 578 | except (TypeError, ValueError): |
| 579 | continue |
| 580 | else: |
| 581 | for idx, col in enumerate(columns): |
| 582 | if not isinstance(col, dict): |
| 583 | continue |
| 584 | field = col.get("field") |
| 585 | if field: |
| 586 | field_to_idx[field] = idx |
| 587 | |
| 588 | data = doc.get("data") or [] |
| 589 | if len(data) == 0: |
| 590 | raise SystemExit(0) |
| 591 | |
| 592 | query_idx = field_to_idx.get("query_text", None) |
| 593 | if query_idx is None: |
| 594 | raise SystemExit("expected no rows, got %s" % len(data)) |
| 595 | |
| 596 | for row in data: |
| 597 | if query_idx >= len(row): |
| 598 | continue |
| 599 | query = str(row[query_idx]).lower() |
| 600 | if "deadlock_a" in query or "deadlock_b" in query: |
| 601 | raise SystemExit("unexpected deadlock rows for test tables, got %s rows" % len(data)) |
| 602 | PY |
| 603 | } |
| 604 | |
| 605 | assert_deadlock_info_error_contains() { |
| 606 | local input="$1" |
| 607 | local expected_status="$2" |
| 608 | local expected_substr="$3" |
| 609 | |
| 610 | if command -v python3 >/dev/null 2>&1; then |
| 611 | python3 - "$input" "$expected_status" "$expected_substr" <<'PY' |
| 612 | import json |
| 613 | import sys |
| 614 | |
| 615 | path = sys.argv[1] |
| 616 | expected_status = int(sys.argv[2]) |
| 617 | expected = sys.argv[3].strip().lower() |
| 618 | with open(path, "r", encoding="utf-8") as fh: |
| 619 | doc = json.load(fh) |
| 620 | |
| 621 | try: |
| 622 | status = int(doc.get("status")) |
| 623 | except (TypeError, ValueError): |
| 624 | raise SystemExit(f"unexpected status value: {doc.get('status')!r}") |
| 625 | |
| 626 | if status != expected_status: |
| 627 | raise SystemExit(f"expected status {expected_status}, got {status}") |
| 628 | |
| 629 | err = str(doc.get("errorMessage") or "").lower() |
| 630 | if expected not in err: |
| 631 | raise SystemExit(f"expected errorMessage to contain {expected!r}, got {err!r}") |
| 632 | PY |
| 633 | return |
| 634 | fi |
| 635 | |
| 636 | python - "$input" "$expected_status" "$expected_substr" <<'PY' |
| 637 | import json |
| 638 | import sys |
| 639 | |
| 640 | path = sys.argv[1] |
| 641 | expected_status = int(sys.argv[2]) |
| 642 | expected = sys.argv[3].strip().lower() |
| 643 | with open(path, "r") as fh: |
| 644 | doc = json.load(fh) |
| 645 | |
| 646 | try: |
| 647 | status = int(doc.get("status")) |
| 648 | except (TypeError, ValueError): |
| 649 | raise SystemExit("unexpected status value: %r" % (doc.get("status"),)) |
| 650 | |
| 651 | if status != expected_status: |
| 652 | raise SystemExit("expected status %s, got %s" % (expected_status, status)) |
| 653 | |
| 654 | err = str(doc.get("errorMessage") or "").lower() |
| 655 | if expected not in err: |
| 656 | raise SystemExit("expected errorMessage to contain %r, got %r" % (expected, err)) |
| 657 | PY |
| 658 | } |
| 659 | |
| 660 | assert_error_info_not_enabled() { |
| 661 | local input="$1" |
| 662 | |
| 663 | if command -v python3 >/dev/null 2>&1; then |
| 664 | python3 - "$input" <<'PY' |
| 665 | import json |
| 666 | import sys |
| 667 | |
| 668 | path = sys.argv[1] |
| 669 | with open(path, "r", encoding="utf-8") as fh: |
| 670 | doc = json.load(fh) |
| 671 | |
| 672 | try: |
| 673 | status = int(doc.get("status")) |
| 674 | except (TypeError, ValueError): |
| 675 | raise SystemExit(f"unexpected status value: {doc.get('status')!r}") |
| 676 | |
| 677 | if status < 400: |
| 678 | raise SystemExit(f"expected error status, got {status}") |
| 679 | |
| 680 | err = str(doc.get("errorMessage") or "").lower() |
| 681 | if "not enabled" not in err: |
| 682 | raise SystemExit(f"expected errorMessage to contain 'not enabled', got {err!r}") |
| 683 | PY |
| 684 | return |
| 685 | fi |
| 686 | |
| 687 | python - "$input" <<'PY' |
| 688 | import json |
| 689 | import sys |
| 690 | |
| 691 | path = sys.argv[1] |
| 692 | with open(path, "r") as fh: |
| 693 | doc = json.load(fh) |
| 694 | |
| 695 | try: |
| 696 | status = int(doc.get("status")) |
| 697 | except (TypeError, ValueError): |
| 698 | raise SystemExit("unexpected status value: %r" % (doc.get("status"),)) |
| 699 | |
| 700 | if status < 400: |
| 701 | raise SystemExit("expected error status, got %s" % status) |
| 702 | |
| 703 | err = str(doc.get("errorMessage") or "").lower() |
| 704 | if "not enabled" not in err: |
| 705 | raise SystemExit("expected errorMessage to contain 'not enabled', got %r" % err) |
| 706 | PY |
| 707 | } |
| 708 | |
| 709 | assert_error_info_has_errors() { |
| 710 | local input="$1" |
| 711 | |
| 712 | if command -v python3 >/dev/null 2>&1; then |
| 713 | python3 - "$input" <<'PY' |
| 714 | import json |
| 715 | import sys |
| 716 | |
| 717 | path = sys.argv[1] |
| 718 | with open(path, "r", encoding="utf-8") as fh: |
| 719 | doc = json.load(fh) |
| 720 | |
| 721 | try: |
| 722 | status = int(doc.get("status")) |
| 723 | except (TypeError, ValueError): |
| 724 | raise SystemExit(f"unexpected status value: {doc.get('status')!r}") |
| 725 | |
| 726 | if status != 200: |
| 727 | raise SystemExit(f"expected status 200, got {status}") |
| 728 | |
| 729 | if doc.get("errorMessage"): |
| 730 | raise SystemExit(f"unexpected errorMessage on status 200: {doc.get('errorMessage')!r}") |
| 731 | |
| 732 | columns = doc.get("columns") or {} |
| 733 | field_to_idx = {} |
| 734 | if isinstance(columns, dict): |
| 735 | for field, col in columns.items(): |
| 736 | if not isinstance(col, dict): |
| 737 | continue |
| 738 | try: |
| 739 | field_to_idx[field] = int(col.get("index")) |
| 740 | except (TypeError, ValueError): |
| 741 | continue |
| 742 | else: |
| 743 | for idx, col in enumerate(columns): |
| 744 | if not isinstance(col, dict): |
| 745 | continue |
| 746 | field = col.get("field") |
| 747 | if field: |
| 748 | field_to_idx[field] = idx |
| 749 | |
| 750 | for required in ("errorNumber", "errorMessage", "query"): |
| 751 | if required not in field_to_idx: |
| 752 | raise SystemExit(f"missing expected column: {required}") |
| 753 | |
| 754 | data = doc.get("data") or [] |
| 755 | if not data: |
| 756 | raise SystemExit("error-info returned no rows") |
| 757 | |
| 758 | num_idx = field_to_idx["errorNumber"] |
| 759 | msg_idx = field_to_idx["errorMessage"] |
| 760 | query_idx = field_to_idx["query"] |
| 761 | |
| 762 | # Error categories to verify: |
| 763 | # 208 - Invalid object name (table not found) |
| 764 | # 102 - Syntax error |
| 765 | # 2627 - Duplicate key / unique constraint violation |
| 766 | # 245 - Data type conversion error |
| 767 | # 8134 - Division by zero |
| 768 | error_categories = { |
| 769 | "table_not_found": {"patterns": ["invalid object name", "netdata_error_map_e2e"], "found": False}, |
| 770 | "syntax_error": {"patterns": ["incorrect syntax", "form"], "found": False}, |
| 771 | "duplicate_key": {"patterns": ["duplicate key", "unique", "primary key", "error_test"], "found": False}, |
| 772 | "data_type": {"patterns": ["conversion failed", "converting"], "found": False}, |
| 773 | "divide_by_zero": {"patterns": ["divide by zero"], "found": False}, |
| 774 | } |
| 775 | |
| 776 | for row in data: |
| 777 | if num_idx >= len(row) or row[num_idx] is None: |
| 778 | continue |
| 779 | msg = str(row[msg_idx]).lower() if msg_idx < len(row) else "" |
| 780 | query = str(row[query_idx]).lower() if query_idx < len(row) else "" |
| 781 | combined = msg + " " + query |
| 782 | for cat, info in error_categories.items(): |
| 783 | if info["found"]: |
| 784 | continue |
| 785 | for pattern in info["patterns"]: |
| 786 | if pattern in combined: |
| 787 | info["found"] = True |
| 788 | break |
| 789 | |
| 790 | missing = [cat for cat, info in error_categories.items() if not info["found"]] |
| 791 | if missing: |
| 792 | raise SystemExit(f"error-info missing error categories: {', '.join(missing)}") |
| 793 | PY |
| 794 | return |
| 795 | fi |
| 796 | |
| 797 | python - "$input" <<'PY' |
| 798 | import json |
| 799 | import sys |
| 800 | |
| 801 | path = sys.argv[1] |
| 802 | with open(path, "r") as fh: |
| 803 | doc = json.load(fh) |
| 804 | |
| 805 | try: |
| 806 | status = int(doc.get("status")) |
| 807 | except (TypeError, ValueError): |
| 808 | raise SystemExit("unexpected status value: %r" % (doc.get("status"),)) |
| 809 | |
| 810 | if status != 200: |
| 811 | raise SystemExit("expected status 200, got %s" % status) |
| 812 | |
| 813 | if doc.get("errorMessage"): |
| 814 | raise SystemExit("unexpected errorMessage on status 200: %r" % (doc.get("errorMessage"),)) |
| 815 | |
| 816 | columns = doc.get("columns") or {} |
| 817 | field_to_idx = {} |
| 818 | if isinstance(columns, dict): |
| 819 | for field, col in columns.items(): |
| 820 | if not isinstance(col, dict): |
| 821 | continue |
| 822 | try: |
| 823 | field_to_idx[field] = int(col.get("index")) |
| 824 | except (TypeError, ValueError): |
| 825 | continue |
| 826 | else: |
| 827 | for idx, col in enumerate(columns): |
| 828 | if not isinstance(col, dict): |
| 829 | continue |
| 830 | field = col.get("field") |
| 831 | if field: |
| 832 | field_to_idx[field] = idx |
| 833 | |
| 834 | for required in ("errorNumber", "errorMessage", "query"): |
| 835 | if required not in field_to_idx: |
| 836 | raise SystemExit("missing expected column: %s" % required) |
| 837 | |
| 838 | data = doc.get("data") or [] |
| 839 | if not data: |
| 840 | raise SystemExit("error-info returned no rows") |
| 841 | |
| 842 | num_idx = field_to_idx["errorNumber"] |
| 843 | msg_idx = field_to_idx["errorMessage"] |
| 844 | query_idx = field_to_idx["query"] |
| 845 | |
| 846 | # Error categories to verify: |
| 847 | # 208 - Invalid object name (table not found) |
| 848 | # 102 - Syntax error |
| 849 | # 2627 - Duplicate key / unique constraint violation |
| 850 | # 245 - Data type conversion error |
| 851 | # 8134 - Division by zero |
| 852 | error_categories = { |
| 853 | "table_not_found": {"patterns": ["invalid object name", "netdata_error_map_e2e"], "found": False}, |
| 854 | "syntax_error": {"patterns": ["incorrect syntax", "form"], "found": False}, |
| 855 | "duplicate_key": {"patterns": ["duplicate key", "unique", "primary key", "error_test"], "found": False}, |
| 856 | "data_type": {"patterns": ["conversion failed", "converting"], "found": False}, |
| 857 | "divide_by_zero": {"patterns": ["divide by zero"], "found": False}, |
| 858 | } |
| 859 | |
| 860 | for row in data: |
| 861 | if num_idx >= len(row) or row[num_idx] is None: |
| 862 | continue |
| 863 | msg = str(row[msg_idx]).lower() if msg_idx < len(row) else "" |
| 864 | query = str(row[query_idx]).lower() if query_idx < len(row) else "" |
| 865 | combined = msg + " " + query |
| 866 | for cat, info in error_categories.items(): |
| 867 | if info["found"]: |
| 868 | continue |
| 869 | for pattern in info["patterns"]: |
| 870 | if pattern in combined: |
| 871 | info["found"] = True |
| 872 | break |
| 873 | |
| 874 | missing = [cat for cat, info in error_categories.items() if not info["found"]] |
| 875 | if missing: |
| 876 | raise SystemExit("error-info missing error categories: %s" % ", ".join(missing)) |
| 877 | PY |
| 878 | } |
| 879 | |
| 880 | assert_top_queries_error_attribution_not_enabled() { |
| 881 | local input="$1" |
| 882 | |
| 883 | if command -v python3 >/dev/null 2>&1; then |
| 884 | python3 - "$input" <<'PY' |
| 885 | import json |
| 886 | import sys |
| 887 | |
| 888 | path = sys.argv[1] |
| 889 | with open(path, "r", encoding="utf-8") as fh: |
| 890 | doc = json.load(fh) |
| 891 | |
| 892 | columns = doc.get("columns") or {} |
| 893 | field_to_idx = {} |
| 894 | if isinstance(columns, dict): |
| 895 | for field, col in columns.items(): |
| 896 | if not isinstance(col, dict): |
| 897 | continue |
| 898 | try: |
| 899 | field_to_idx[field] = int(col.get("index")) |
| 900 | except (TypeError, ValueError): |
| 901 | continue |
| 902 | else: |
| 903 | for idx, col in enumerate(columns): |
| 904 | if not isinstance(col, dict): |
| 905 | continue |
| 906 | field = col.get("field") |
| 907 | if field: |
| 908 | field_to_idx[field] = idx |
| 909 | |
| 910 | if "errorAttribution" not in field_to_idx: |
| 911 | raise SystemExit("missing expected column: errorAttribution") |
| 912 | |
| 913 | data = doc.get("data") or [] |
| 914 | idx = field_to_idx["errorAttribution"] |
| 915 | for row in data: |
| 916 | if idx >= len(row): |
| 917 | continue |
| 918 | if str(row[idx]) != "not_enabled": |
| 919 | raise SystemExit(f"expected errorAttribution 'not_enabled', got {row[idx]!r}") |
| 920 | PY |
| 921 | return |
| 922 | fi |
| 923 | |
| 924 | python - "$input" <<'PY' |
| 925 | import json |
| 926 | import sys |
| 927 | |
| 928 | path = sys.argv[1] |
| 929 | with open(path, "r") as fh: |
| 930 | doc = json.load(fh) |
| 931 | |
| 932 | columns = doc.get("columns") or {} |
| 933 | field_to_idx = {} |
| 934 | if isinstance(columns, dict): |
| 935 | for field, col in columns.items(): |
| 936 | if not isinstance(col, dict): |
| 937 | continue |
| 938 | try: |
| 939 | field_to_idx[field] = int(col.get("index")) |
| 940 | except (TypeError, ValueError): |
| 941 | continue |
| 942 | else: |
| 943 | for idx, col in enumerate(columns): |
| 944 | if not isinstance(col, dict): |
| 945 | continue |
| 946 | field = col.get("field") |
| 947 | if field: |
| 948 | field_to_idx[field] = idx |
| 949 | |
| 950 | if "errorAttribution" not in field_to_idx: |
| 951 | raise SystemExit("missing expected column: errorAttribution") |
| 952 | |
| 953 | data = doc.get("data") or [] |
| 954 | idx = field_to_idx["errorAttribution"] |
| 955 | for row in data: |
| 956 | if idx >= len(row): |
| 957 | continue |
| 958 | if str(row[idx]) != "not_enabled": |
| 959 | raise SystemExit("expected errorAttribution 'not_enabled', got %r" % row[idx]) |
| 960 | PY |
| 961 | } |
| 962 | |
| 963 | assert_top_queries_error_attribution_active() { |
| 964 | local input="$1" |
| 965 | |
| 966 | if command -v python3 >/dev/null 2>&1; then |
| 967 | python3 - "$input" <<'PY' |
| 968 | import json |
| 969 | import sys |
| 970 | |
| 971 | path = sys.argv[1] |
| 972 | with open(path, "r", encoding="utf-8") as fh: |
| 973 | doc = json.load(fh) |
| 974 | |
| 975 | columns = doc.get("columns") or {} |
| 976 | field_to_idx = {} |
| 977 | if isinstance(columns, dict): |
| 978 | for field, col in columns.items(): |
| 979 | if not isinstance(col, dict): |
| 980 | continue |
| 981 | try: |
| 982 | field_to_idx[field] = int(col.get("index")) |
| 983 | except (TypeError, ValueError): |
| 984 | continue |
| 985 | else: |
| 986 | for idx, col in enumerate(columns): |
| 987 | if not isinstance(col, dict): |
| 988 | continue |
| 989 | field = col.get("field") |
| 990 | if field: |
| 991 | field_to_idx[field] = idx |
| 992 | |
| 993 | for required in ("errorAttribution",): |
| 994 | if required not in field_to_idx: |
| 995 | raise SystemExit(f"missing expected column: {required}") |
| 996 | |
| 997 | data = doc.get("data") or [] |
| 998 | status_idx = field_to_idx["errorAttribution"] |
| 999 | |
| 1000 | for row in data: |
| 1001 | if status_idx >= len(row): |
| 1002 | continue |
| 1003 | status = str(row[status_idx]) |
| 1004 | if status not in ("enabled", "no_data"): |
| 1005 | raise SystemExit(f"unexpected errorAttribution status {status!r}") |
| 1006 | PY |
| 1007 | return |
| 1008 | fi |
| 1009 | |
| 1010 | python - "$input" <<'PY' |
| 1011 | import json |
| 1012 | import sys |
| 1013 | |
| 1014 | path = sys.argv[1] |
| 1015 | with open(path, "r") as fh: |
| 1016 | doc = json.load(fh) |
| 1017 | |
| 1018 | columns = doc.get("columns") or {} |
| 1019 | field_to_idx = {} |
| 1020 | if isinstance(columns, dict): |
| 1021 | for field, col in columns.items(): |
| 1022 | if not isinstance(col, dict): |
| 1023 | continue |
| 1024 | try: |
| 1025 | field_to_idx[field] = int(col.get("index")) |
| 1026 | except (TypeError, ValueError): |
| 1027 | continue |
| 1028 | else: |
| 1029 | for idx, col in enumerate(columns): |
| 1030 | if not isinstance(col, dict): |
| 1031 | continue |
| 1032 | field = col.get("field") |
| 1033 | if field: |
| 1034 | field_to_idx[field] = idx |
| 1035 | |
| 1036 | for required in ("errorAttribution",): |
| 1037 | if required not in field_to_idx: |
| 1038 | raise SystemExit("missing expected column: %s" % required) |
| 1039 | |
| 1040 | data = doc.get("data") or [] |
| 1041 | status_idx = field_to_idx["errorAttribution"] |
| 1042 | |
| 1043 | for row in data: |
| 1044 | if status_idx >= len(row): |
| 1045 | continue |
| 1046 | status = str(row[status_idx]) |
| 1047 | if status not in ("enabled", "no_data"): |
| 1048 | raise SystemExit("unexpected errorAttribution status %r" % status) |
| 1049 | PY |
| 1050 | } |
| 1051 | |
| 1052 | assert_top_queries_error_attribution_mapped() { |
| 1053 | local top_queries="$1" |
| 1054 | local error_info="$2" |
| 1055 | |
| 1056 | if command -v python3 >/dev/null 2>&1; then |
| 1057 | python3 - "$top_queries" "$error_info" <<'PY' |
| 1058 | import json |
| 1059 | import sys |
| 1060 | |
| 1061 | top_path = sys.argv[1] |
| 1062 | err_path = sys.argv[2] |
| 1063 | |
| 1064 | with open(err_path, "r", encoding="utf-8") as fh: |
| 1065 | err_doc = json.load(fh) |
| 1066 | |
| 1067 | err_cols = err_doc.get("columns") or {} |
| 1068 | err_idx = {} |
| 1069 | if isinstance(err_cols, dict): |
| 1070 | for field, col in err_cols.items(): |
| 1071 | if not isinstance(col, dict): |
| 1072 | continue |
| 1073 | try: |
| 1074 | err_idx[field] = int(col.get("index")) |
| 1075 | except (TypeError, ValueError): |
| 1076 | continue |
| 1077 | else: |
| 1078 | for idx, col in enumerate(err_cols): |
| 1079 | if not isinstance(col, dict): |
| 1080 | continue |
| 1081 | field = col.get("field") |
| 1082 | if field: |
| 1083 | err_idx[field] = idx |
| 1084 | |
| 1085 | for required in ("errorMessage", "errorNumber", "query", "queryHash"): |
| 1086 | if required not in err_idx: |
| 1087 | raise SystemExit(f"missing expected error-info column: {required}") |
| 1088 | |
| 1089 | def normalize(text: str) -> str: |
| 1090 | return " ".join(text.split()).strip().rstrip(";").strip() |
| 1091 | |
| 1092 | error_rows = err_doc.get("data") or [] |
| 1093 | candidates = [] |
| 1094 | for row in error_rows: |
| 1095 | msg = str(row[err_idx["errorMessage"]]).lower() if err_idx["errorMessage"] < len(row) else "" |
| 1096 | err_no = row[err_idx["errorNumber"]] if err_idx["errorNumber"] < len(row) else None |
| 1097 | query = str(row[err_idx["query"]]).lower() if err_idx["query"] < len(row) else "" |
| 1098 | qh = row[err_idx["queryHash"]] if err_idx["queryHash"] < len(row) else None |
| 1099 | try: |
| 1100 | err_no_val = int(err_no) |
| 1101 | except Exception: |
| 1102 | continue |
| 1103 | if err_no_val != 208: |
| 1104 | continue |
| 1105 | if "invalid object name" in msg and "netdata_error_map_e2e" in query: |
| 1106 | candidates.append((str(qh) if qh else "", normalize(query))) |
| 1107 | |
| 1108 | if not candidates: |
| 1109 | raise SystemExit("no error-info row contained invalid object name for netdata_error_map_e2e") |
| 1110 | |
| 1111 | with open(top_path, "r", encoding="utf-8") as fh: |
| 1112 | doc = json.load(fh) |
| 1113 | |
| 1114 | columns = doc.get("columns") or {} |
| 1115 | field_to_idx = {} |
| 1116 | if isinstance(columns, dict): |
| 1117 | for field, col in columns.items(): |
| 1118 | if not isinstance(col, dict): |
| 1119 | continue |
| 1120 | try: |
| 1121 | field_to_idx[field] = int(col.get("index")) |
| 1122 | except (TypeError, ValueError): |
| 1123 | continue |
| 1124 | else: |
| 1125 | for idx, col in enumerate(columns): |
| 1126 | if not isinstance(col, dict): |
| 1127 | continue |
| 1128 | field = col.get("field") |
| 1129 | if field: |
| 1130 | field_to_idx[field] = idx |
| 1131 | |
| 1132 | for required in ("query", "queryHash", "errorAttribution", "errorNumber", "errorMessage"): |
| 1133 | if required not in field_to_idx: |
| 1134 | raise SystemExit(f"missing expected column: {required}") |
| 1135 | |
| 1136 | data = doc.get("data") or [] |
| 1137 | status_idx = field_to_idx["errorAttribution"] |
| 1138 | num_idx = field_to_idx["errorNumber"] |
| 1139 | msg_idx = field_to_idx["errorMessage"] |
| 1140 | hash_idx = field_to_idx["queryHash"] |
| 1141 | |
| 1142 | matched = False |
| 1143 | for row in data: |
| 1144 | if status_idx >= len(row): |
| 1145 | continue |
| 1146 | if hash_idx >= len(row): |
| 1147 | continue |
| 1148 | status = str(row[status_idx]) if status_idx < len(row) else "" |
| 1149 | if status != "enabled": |
| 1150 | continue |
| 1151 | err_no = row[num_idx] if num_idx < len(row) else None |
| 1152 | try: |
| 1153 | err_no_val = int(err_no) |
| 1154 | except Exception: |
| 1155 | continue |
| 1156 | if err_no_val != 208: |
| 1157 | continue |
| 1158 | msg = str(row[msg_idx]).lower() if msg_idx < len(row) and row[msg_idx] is not None else "" |
| 1159 | if "invalid object name" not in msg: |
| 1160 | continue |
| 1161 | row_hash = str(row[hash_idx]) if hash_idx < len(row) and row[hash_idx] is not None else "" |
| 1162 | row_query = normalize(str(row[field_to_idx["query"]]).lower()) if field_to_idx["query"] < len(row) else "" |
| 1163 | for cand_hash, cand_query in candidates: |
| 1164 | if cand_hash and row_hash == cand_hash: |
| 1165 | matched = True |
| 1166 | break |
| 1167 | if cand_query and row_query == cand_query: |
| 1168 | matched = True |
| 1169 | break |
| 1170 | if matched: |
| 1171 | break |
| 1172 | |
| 1173 | if not matched: |
| 1174 | raise SystemExit("no top-queries row had enabled error attribution for netdata_error_map_e2e") |
| 1175 | PY |
| 1176 | return |
| 1177 | fi |
| 1178 | |
| 1179 | python - "$top_queries" "$error_info" <<'PY' |
| 1180 | import json |
| 1181 | import sys |
| 1182 | |
| 1183 | top_path = sys.argv[1] |
| 1184 | err_path = sys.argv[2] |
| 1185 | |
| 1186 | with open(err_path, "r") as fh: |
| 1187 | err_doc = json.load(fh) |
| 1188 | |
| 1189 | err_cols = err_doc.get("columns") or {} |
| 1190 | err_idx = {} |
| 1191 | if isinstance(err_cols, dict): |
| 1192 | for field, col in err_cols.items(): |
| 1193 | if not isinstance(col, dict): |
| 1194 | continue |
| 1195 | try: |
| 1196 | err_idx[field] = int(col.get("index")) |
| 1197 | except (TypeError, ValueError): |
| 1198 | continue |
| 1199 | else: |
| 1200 | for idx, col in enumerate(err_cols): |
| 1201 | if not isinstance(col, dict): |
| 1202 | continue |
| 1203 | field = col.get("field") |
| 1204 | if field: |
| 1205 | err_idx[field] = idx |
| 1206 | |
| 1207 | for required in ("errorMessage", "errorNumber", "query", "queryHash"): |
| 1208 | if required not in err_idx: |
| 1209 | raise SystemExit("missing expected error-info column: %s" % required) |
| 1210 | |
| 1211 | def normalize(text): |
| 1212 | return " ".join(text.split()).strip().rstrip(";").strip() |
| 1213 | |
| 1214 | error_rows = err_doc.get("data") or [] |
| 1215 | candidates = [] |
| 1216 | for row in error_rows: |
| 1217 | msg = str(row[err_idx["errorMessage"]]).lower() if err_idx["errorMessage"] < len(row) else "" |
| 1218 | err_no = row[err_idx["errorNumber"]] if err_idx["errorNumber"] < len(row) else None |
| 1219 | query = str(row[err_idx["query"]]).lower() if err_idx["query"] < len(row) else "" |
| 1220 | qh = row[err_idx["queryHash"]] if err_idx["queryHash"] < len(row) else None |
| 1221 | try: |
| 1222 | err_no_val = int(err_no) |
| 1223 | except Exception: |
| 1224 | continue |
| 1225 | if err_no_val != 208: |
| 1226 | continue |
| 1227 | if "invalid object name" in msg and "netdata_error_map_e2e" in query: |
| 1228 | candidates.append((str(qh) if qh else "", normalize(query))) |
| 1229 | |
| 1230 | if not candidates: |
| 1231 | raise SystemExit("no error-info row contained invalid object name for netdata_error_map_e2e") |
| 1232 | |
| 1233 | with open(top_path, "r") as fh: |
| 1234 | doc = json.load(fh) |
| 1235 | |
| 1236 | columns = doc.get("columns") or {} |
| 1237 | field_to_idx = {} |
| 1238 | if isinstance(columns, dict): |
| 1239 | for field, col in columns.items(): |
| 1240 | if not isinstance(col, dict): |
| 1241 | continue |
| 1242 | try: |
| 1243 | field_to_idx[field] = int(col.get("index")) |
| 1244 | except (TypeError, ValueError): |
| 1245 | continue |
| 1246 | else: |
| 1247 | for idx, col in enumerate(columns): |
| 1248 | if not isinstance(col, dict): |
| 1249 | continue |
| 1250 | field = col.get("field") |
| 1251 | if field: |
| 1252 | field_to_idx[field] = idx |
| 1253 | |
| 1254 | for required in ("query", "queryHash", "errorAttribution", "errorNumber", "errorMessage"): |
| 1255 | if required not in field_to_idx: |
| 1256 | raise SystemExit("missing expected column: %s" % required) |
| 1257 | |
| 1258 | data = doc.get("data") or [] |
| 1259 | status_idx = field_to_idx["errorAttribution"] |
| 1260 | num_idx = field_to_idx["errorNumber"] |
| 1261 | msg_idx = field_to_idx["errorMessage"] |
| 1262 | hash_idx = field_to_idx["queryHash"] |
| 1263 | |
| 1264 | matched = False |
| 1265 | for row in data: |
| 1266 | if status_idx >= len(row): |
| 1267 | continue |
| 1268 | if hash_idx >= len(row): |
| 1269 | continue |
| 1270 | status = str(row[status_idx]) if status_idx < len(row) else "" |
| 1271 | if status != "enabled": |
| 1272 | continue |
| 1273 | err_no = row[num_idx] if num_idx < len(row) else None |
| 1274 | try: |
| 1275 | err_no_val = int(err_no) |
| 1276 | except Exception: |
| 1277 | continue |
| 1278 | if err_no_val != 208: |
| 1279 | continue |
| 1280 | msg = str(row[msg_idx]).lower() if msg_idx < len(row) and row[msg_idx] is not None else "" |
| 1281 | if "invalid object name" not in msg: |
| 1282 | continue |
| 1283 | row_hash = str(row[hash_idx]) if hash_idx < len(row) and row[hash_idx] is not None else "" |
| 1284 | row_query = normalize(str(row[field_to_idx["query"]]).lower()) if field_to_idx["query"] < len(row) else "" |
| 1285 | for cand_hash, cand_query in candidates: |
| 1286 | if cand_hash and row_hash == cand_hash: |
| 1287 | matched = True |
| 1288 | break |
| 1289 | if cand_query and row_query == cand_query: |
| 1290 | matched = True |
| 1291 | break |
| 1292 | if matched: |
| 1293 | break |
| 1294 | |
| 1295 | if not matched: |
| 1296 | raise SystemExit("no top-queries row had enabled error attribution for netdata_error_map_e2e") |
| 1297 | PY |
| 1298 | } |
| 1299 | |
| 1300 | assert_top_queries_plan_ops() { |
| 1301 | local input="$1" |
| 1302 | |
| 1303 | if command -v python3 >/dev/null 2>&1; then |
| 1304 | python3 - "$input" <<'PY' |
| 1305 | import json |
| 1306 | import sys |
| 1307 | |
| 1308 | path = sys.argv[1] |
| 1309 | with open(path, "r", encoding="utf-8") as fh: |
| 1310 | doc = json.load(fh) |
| 1311 | |
| 1312 | columns = doc.get("columns") or {} |
| 1313 | field_to_idx = {} |
| 1314 | if isinstance(columns, dict): |
| 1315 | for field, col in columns.items(): |
| 1316 | if not isinstance(col, dict): |
| 1317 | continue |
| 1318 | try: |
| 1319 | field_to_idx[field] = int(col.get("index")) |
| 1320 | except (TypeError, ValueError): |
| 1321 | continue |
| 1322 | else: |
| 1323 | for idx, col in enumerate(columns): |
| 1324 | if not isinstance(col, dict): |
| 1325 | continue |
| 1326 | field = col.get("field") |
| 1327 | if field: |
| 1328 | field_to_idx[field] = idx |
| 1329 | |
| 1330 | for required in ("query", "hashMatch", "sorts"): |
| 1331 | if required not in field_to_idx: |
| 1332 | raise SystemExit(f"missing expected column: {required}") |
| 1333 | |
| 1334 | data = doc.get("data") or [] |
| 1335 | query_idx = field_to_idx["query"] |
| 1336 | hash_idx = field_to_idx["hashMatch"] |
| 1337 | sort_idx = field_to_idx["sorts"] |
| 1338 | |
| 1339 | matched = False |
| 1340 | for row in data: |
| 1341 | if query_idx >= len(row): |
| 1342 | continue |
| 1343 | query = str(row[query_idx]).lower() |
| 1344 | if "join" not in query or "sample" not in query: |
| 1345 | continue |
| 1346 | hash_val = row[hash_idx] if hash_idx < len(row) else 0 |
| 1347 | sort_val = row[sort_idx] if sort_idx < len(row) else 0 |
| 1348 | try: |
| 1349 | hash_val = int(hash_val) |
| 1350 | except Exception: |
| 1351 | hash_val = 0 |
| 1352 | try: |
| 1353 | sort_val = int(sort_val) |
| 1354 | except Exception: |
| 1355 | sort_val = 0 |
| 1356 | if hash_val > 0 and sort_val > 0: |
| 1357 | matched = True |
| 1358 | break |
| 1359 | |
| 1360 | if not matched: |
| 1361 | raise SystemExit("no top-queries row had hashMatch and sorts counts for the join query") |
| 1362 | PY |
| 1363 | return |
| 1364 | fi |
| 1365 | |
| 1366 | python - "$input" <<'PY' |
| 1367 | import json |
| 1368 | import sys |
| 1369 | |
| 1370 | path = sys.argv[1] |
| 1371 | with open(path, "r") as fh: |
| 1372 | doc = json.load(fh) |
| 1373 | |
| 1374 | columns = doc.get("columns") or {} |
| 1375 | field_to_idx = {} |
| 1376 | if isinstance(columns, dict): |
| 1377 | for field, col in columns.items(): |
| 1378 | if not isinstance(col, dict): |
| 1379 | continue |
| 1380 | try: |
| 1381 | field_to_idx[field] = int(col.get("index")) |
| 1382 | except (TypeError, ValueError): |
| 1383 | continue |
| 1384 | else: |
| 1385 | for idx, col in enumerate(columns): |
| 1386 | if not isinstance(col, dict): |
| 1387 | continue |
| 1388 | field = col.get("field") |
| 1389 | if field: |
| 1390 | field_to_idx[field] = idx |
| 1391 | |
| 1392 | for required in ("query", "hashMatch", "sorts"): |
| 1393 | if required not in field_to_idx: |
| 1394 | raise SystemExit("missing expected column: %s" % required) |
| 1395 | |
| 1396 | data = doc.get("data") or [] |
| 1397 | query_idx = field_to_idx["query"] |
| 1398 | hash_idx = field_to_idx["hashMatch"] |
| 1399 | sort_idx = field_to_idx["sorts"] |
| 1400 | |
| 1401 | matched = False |
| 1402 | for row in data: |
| 1403 | if query_idx >= len(row): |
| 1404 | continue |
| 1405 | query = str(row[query_idx]).lower() |
| 1406 | if "join" not in query or "sample" not in query: |
| 1407 | continue |
| 1408 | hash_val = row[hash_idx] if hash_idx < len(row) else 0 |
| 1409 | sort_val = row[sort_idx] if sort_idx < len(row) else 0 |
| 1410 | try: |
| 1411 | hash_val = int(hash_val) |
| 1412 | except Exception: |
| 1413 | hash_val = 0 |
| 1414 | try: |
| 1415 | sort_val = int(sort_val) |
| 1416 | except Exception: |
| 1417 | sort_val = 0 |
| 1418 | if hash_val > 0 and sort_val > 0: |
| 1419 | matched = True |
| 1420 | break |
| 1421 | |
| 1422 | if not matched: |
| 1423 | raise SystemExit("no top-queries row had hashMatch and sorts counts for the join query") |
| 1424 | PY |
| 1425 | } |
| 1426 | |
| 1427 | verify_deadlock_info_no_deadlock() { |
| 1428 | local output |
| 1429 | |
| 1430 | output="$(run_mssql_function_with_retry deadlock-info '__job:local' 'false')" |
| 1431 | validate "$output" |
| 1432 | assert_deadlock_info_empty_success "$output" |
| 1433 | } |
| 1434 | |
| 1435 | verify_deadlock_info() { |
| 1436 | local attempt |
| 1437 | local output |
| 1438 | local found="false" |
| 1439 | |
| 1440 | for attempt in 1 2 3 4 5; do |
| 1441 | induce_deadlock_once |
| 1442 | output="$(run_mssql_function_with_retry deadlock-info '__job:local' 'false')" |
| 1443 | if has_min_rows "$output" 1; then |
| 1444 | validate "$output" --min-rows 1 |
| 1445 | if assert_deadlock_info_content "$output"; then |
| 1446 | found="true" |
| 1447 | break |
| 1448 | fi |
| 1449 | fi |
| 1450 | sleep 1 |
| 1451 | done |
| 1452 | |
| 1453 | if [ "$found" != "true" ]; then |
| 1454 | echo "deadlock-info did not produce valid deadlock attribution after 5 attempts" >&2 |
| 1455 | return 1 |
| 1456 | fi |
| 1457 | |
| 1458 | # Verify column visibility rules |
| 1459 | assert_column_visibility "$output" "deadlock-info" |
| 1460 | } |
| 1461 | |
| 1462 | verify_deadlock_info_no_deadlock |
| 1463 | verify_deadlock_info |
| 1464 | |
| 1465 | assert_top_queries_error_attribution_not_enabled "$WORKDIR/mssql-top-queries.json" |
| 1466 | |
| 1467 | error_output="$(run_mssql_function_with_retry error-info '__job:local' 'false')" |
| 1468 | assert_error_info_not_enabled "$error_output" |
| 1469 | |
| 1470 | mssql_exec_sa "IF EXISTS (SELECT 1 FROM sys.server_event_sessions WHERE name = 'netdata_errors') DROP EVENT SESSION [netdata_errors] ON SERVER;" |
| 1471 | mssql_exec_sa "CREATE EVENT SESSION [netdata_errors] ON SERVER ADD EVENT sqlserver.error_reported(ACTION(sqlserver.sql_text, sqlserver.query_hash)) ADD TARGET package0.ring_buffer;" |
| 1472 | mssql_exec_sa "ALTER EVENT SESSION [netdata_errors] ON SERVER STATE = START;" |
| 1473 | mssql_exec_sa "ALTER DATABASE netdata SET QUERY_STORE (QUERY_CAPTURE_MODE = ALL, OPERATION_MODE = READ_WRITE);" |
| 1474 | |
| 1475 | mssql_exec_sa "IF OBJECT_ID('dbo.netdata_error_map_e2e', 'U') IS NOT NULL DROP TABLE dbo.netdata_error_map_e2e;" |
| 1476 | mssql_exec_sa "CREATE TABLE dbo.netdata_error_map_e2e (id int NOT NULL PRIMARY KEY);" |
| 1477 | mssql_exec_sa "INSERT INTO dbo.netdata_error_map_e2e (id) VALUES (1), (2), (3);" |
| 1478 | |
| 1479 | for _ in 1 2 3 4 5 6 7 8 9 10; do |
| 1480 | mssql_exec_sa "SELECT COUNT(*) FROM dbo.netdata_error_map_e2e;" |
| 1481 | done |
| 1482 | |
| 1483 | mssql_exec_sa "DROP TABLE dbo.netdata_error_map_e2e;" |
| 1484 | |
| 1485 | # Generate errors for multiple categories: |
| 1486 | # 1. Table not found (error 208) |
| 1487 | for _ in 1 2 3; do |
| 1488 | mssql_exec_sa_allow_error "SELECT COUNT(*) FROM dbo.netdata_error_map_e2e;" |
| 1489 | done |
| 1490 | |
| 1491 | # 2. Syntax error (error 102) |
| 1492 | for _ in 1 2 3; do |
| 1493 | mssql_exec_sa_allow_error "SELECT * FORM dbo.sample;" |
| 1494 | done |
| 1495 | |
| 1496 | # 3. Duplicate key / constraint violation (error 2627) |
| 1497 | for _ in 1 2 3; do |
| 1498 | mssql_exec_sa_allow_error "INSERT INTO dbo.error_test (id, unique_col, int_col) VALUES (1, 'new_value', 200);" |
| 1499 | mssql_exec_sa_allow_error "INSERT INTO dbo.error_test (id, unique_col, int_col) VALUES (99, 'existing_value', 300);" |
| 1500 | done |
| 1501 | |
| 1502 | # 4. Data type conversion error (error 245) |
| 1503 | for _ in 1 2 3; do |
| 1504 | mssql_exec_sa_allow_error "SELECT CAST('not_a_number' AS INT);" |
| 1505 | done |
| 1506 | |
| 1507 | # 5. Division by zero (error 8134) |
| 1508 | for _ in 1 2 3; do |
| 1509 | mssql_exec_sa_allow_error "SELECT 1/0;" |
| 1510 | done |
| 1511 | |
| 1512 | for _ in 1 2 3 4 5; do |
| 1513 | mssql_exec_sa "SET NOCOUNT ON; SELECT a.id, b.name FROM dbo.sample a JOIN dbo.sample b ON a.id = b.id ORDER BY a.value + b.value DESC OPTION (HASH JOIN);" |
| 1514 | done |
| 1515 | |
| 1516 | mssql_exec_sa "EXEC sys.sp_query_store_flush_db;" |
| 1517 | sleep 2 |
| 1518 | |
| 1519 | error_output="$(run_mssql_function_with_retry error-info '__job:local' 'true')" |
| 1520 | assert_error_info_has_errors "$error_output" |
| 1521 | assert_column_visibility "$error_output" "error-info" |
| 1522 | assert_unique_key_populated "$error_output" "error-info" |
| 1523 | |
| 1524 | run_mssql_top_queries_with_retry |
| 1525 | assert_top_queries_error_attribution_active "$WORKDIR/mssql-top-queries.json" |
| 1526 | assert_top_queries_error_attribution_mapped "$WORKDIR/mssql-top-queries.json" "$WORKDIR/mssql-error-info.json" |
| 1527 | assert_top_queries_plan_ops "$WORKDIR/mssql-top-queries.json" |
| 1528 | assert_column_visibility "$WORKDIR/mssql-top-queries.json" "top-queries" |
| 1529 | |
| 1530 | echo "E2E checks passed for ${MSSQL_VARIANT_LABEL}." >&2 |