ImportWAL replays WAL entries into a running IoTDB instance. It supports tree-model inserts and deletes, table-model inserts, table-model deletes that can be converted to SQL, and memtable snapshots. WAL files are read in version order within each parent directory. When several WAL directories are supplied, the directories can be replayed in parallel.
tools/import-wal.sh launcher on Linux or macOS, or tools/windows/import-wal.bat on Windows, from an IoTDB distribution.JAVA_HOME; the shell launcher uses JAVA_HOME when it is set and otherwise looks for java on PATH.tools/import-wal.sh --file <wal-file-or-directory> [options]
tools\windows\import-wal.bat --file <wal-file-or-directory> [options]
--file may name one .wal file or a directory. A directory is searched recursively for files with the .wal suffix (case-insensitive). Files are grouped by their parent directory and sorted by WAL version, then by file name.
Run tools/import-wal.sh --help (or the Windows equivalent) to print the same option list as the current binary.
| Option | Default | Description |
|---|---|---|
-f, --file <path> | required | A WAL file or a directory containing WAL files. |
-h, --host <host> | 127.0.0.1 | Target IoTDB host. |
-p, --port <port> | 6667 | Target IoTDB RPC port. |
-u, --username <name> | root | Target IoTDB user name. |
-pw, --password <password> | prompted | Target IoTDB password. It is read from the terminal when omitted. |
-db, --database <name> | inferred when possible | Target table-model database. An explicit value applies to every WAL directory and takes precedence over inference. |
--skip_db_confirmation | off | Accept every inferred table-model database without prompting. The explicit --database value still takes precedence. |
-os, --on_success <none|delete> | none | Keep source WAL files, or delete files that were completely replayed after the import succeeds. Files with skipped entries or corruption are retained. |
-tn, --thread_num <count> | 1 | Number of WAL directories to replay concurrently. Files in one directory remain ordered. |
--on_delete <ask|execute|skip|terminate> | ask | Policy for tree-model and table-model delete entries. |
--on_object <ask|skip|terminate> | ask | Policy for ObjectNode entries. execute is not supported. |
--on_unsupported <ask|skip|terminate> | ask | Policy for unsupported entries, including table deletes that cannot be converted to SQL. |
--on_corrupted <ask|skip|terminate> | ask | Policy for a truncated or corrupted WAL file. Skipping abandons the rest of that file. |
--help | Print command-line help. |
For unattended imports, specify a policy for every operation that may require a decision. An ask policy without an interactive terminal terminates the operation (or fails for a corrupted file), so automation should use execute, skip, or terminate as appropriate.
When --database is omitted, the tool examines each WAL file's immediate parent directory. IoTDB WAL directories are normally named <database>-<regionId>, for example:
wal/ factory-0/ wal-00000000000000000001.wal factory-east-2/ wal-00000000000000000002.wal
The final -<regionId> is removed, so the inferred databases are factory and factory-east. The database name must satisfy IoTDB's table-model database naming rules. Directories that do not match the pattern are left without an inferred database; table-model entries from such a directory require --database.
For every inferred directory, an interactive run asks whether to replay into that database:
... [y] yes, [a] accept all inferred databases, [N] quit:
Enter a or all to accept the remaining inferred database names for this import. Each directory still uses its own inferred database. Use --skip_db_confirmation to suppress these prompts.
With the default ask policies, the following responses are available:
e/execute, s/skip, a/all to execute all deletes for the current data model, l/skip_all to skip all deletes for the current data model, or q/quit to stop.ObjectNode and other unsupported entries: s/skip, l/skip_all, or q/quit.s/skip to skip the current file, l/skip_all to skip all corrupted files, or q/quit to stop.Already replayed operations are not rolled back when a corrupted file is skipped. Skipped entries and skipped corrupted files are reported in the final summary and their source files are retained.
Replay a tree-model WAL directory and keep the source files:
tools/import-wal.sh --file /backup/iotdb/wal/root-0 --username root --password secret
Replay table-model WAL directories using the database names inferred from their directory names:
tools/import-wal.sh \ --file /backup/iotdb/wal \ --password secret \ --skip_db_confirmation \ --on_delete skip \ --on_object skip \ --on_unsupported skip \ --on_corrupted skip \ --thread_num 4
Run a non-interactive import with one explicit target database and delete only fully replayed WAL files:
tools/import-wal.sh \ --file /backup/iotdb/wal/factory-0 \ --database factory \ --password secret \ --on_delete execute \ --on_object skip \ --on_unsupported terminate \ --on_corrupted terminate \ --on_success delete
The process exits with 0 after a successful import and 1 when argument parsing, replay, or source-file handling fails.
tools/export-data.sh (Windows: tools/windows/export-data.bat) exports Tree or Table results as CSV, SQL or TsFile. Run -help csv, -help sql or -help tsfile for the format's options.
tools/export-data.sh -h 127.0.0.1 -p 6667 -u root -pw secret \ -ft csv -t ./export -mf ./measurements.txt \ -start_time 2026-10-14T00:00:00+08:00 -end_time 2026-10-14T23:59:59+08:00
-mf/--measurement_file accepts a UTF-8 file containing one IoTDB path pattern per line:
# Exact paths and IoTDB patterns can be combined. root.factory.device01.temperature root.factory.*.pre* root.factory.**
A first-line BOM, blank lines, surrounding whitespace and lines beginning with # are accepted. The entire file is validated before any server query. Each distinct pattern is resolved with SHOW TIMESERIES; *, ** and pre* use IoTDB path syntax. Actual measurements are merged and deduplicated in first-discovery order, then exported sequentially. Each unmatched pattern produces a warning that names the pattern and its line, such as WARNING: No measurements matched root.factory.missing.* (measurement file line 5).; the log also records the source file. The command fails if no actual measurements remain. -mf is available only in Tree mode and cannot be combined with -q.
Each measurement uses its actual full path as the output filename. CSV and SQL are split by -lpf/--lines_per_file (default 10,000 rows); TsFile produces one file per measurement:
export/root.factory.device01.temperature_0.csv export/root.factory.device01.temperature_1.csv export/root.factory.device01.temperature.tsfile
The existing -pfn/--prefix_file_name supplies an optional extra prefix in measurement-file mode: -pfn backup produces backup_root.factory.device01.temperature_0.csv. No default dump prefix is added in this mode, and -f is not an alias. Query and Table exports retain names such as dump0_0.csv, dump0_0.sql and dump0.tsfile. Characters that are unsafe in filenames are encoded as %HH. Oversized filenames and case/normalization collisions fail before exporting data.
-start_time and -end_time form an inclusive time interval. Values may be integer timestamps in the server‘s precision or date-time strings interpreted in the session’s timezone (-tz). A single bound is supported; a reversed or invalid interval fails before data queries. These bounds also apply to automatic Table exports. Queries, whether given with -q or at the Tree prompt, run as written: the bounds are ignored and reported as a WARNING. A measurement with no rows in the interval still counts as completed; CSV/SQL create no data file for it. SQL exports contain INSERT statements for reimport into compatible existing schema; TsFile retains measurement types, encoding, compression and device alignment.
tools/export-data.sh -ft csv -t ./export -q 'SELECT * FROM root.factory.**' tools/export-data.sh -sql_dialect table -db factory -table readings \ -ft tsfile -t ./export -start_time 100 -end_time 200 tools/export-data.sh -sql_dialect table -db factory \ -ft sql -t ./export -q 'SELECT * FROM readings WHERE time >= 100'
Tree mode prompts for a query when neither -mf nor -q is supplied. Table mode automatically exports the selected table or all tables in -db. Each query in -q writes its own files (dump0_0.csv, dump1_0.csv, ...). A Table query exported as SQL or TsFile is written back to the table it reads, so it must select * or column names from a single table of -db: aliases, expressions, FILL, joins, subqueries and CTEs would write values that the table does not hold. The query must return the time column and at least one FIELD column, and TsFile queries must include every TAG column to preserve device identity. All queries are checked against their tables before the first one runs. Like the time bounds, -table does not apply to queries and is reported as a WARNING.
SQL output quotes strings and dates in single quotes, doubling a quote inside them. A Tree query row that spans several devices becomes one SQL INSERT per device that has values in the row, and columns without a value are left out. These statements stay in the same file and count as one data row for -lpf, exported_data_rows and file_rows. ALIGN BY DEVICE and Table rows write NULL for a missing value. With -dt true, CSV headers in both models add each value column's type, such as s1(DOUBLE).
Every invocation writes a UTF-8 append log and reports progress every two seconds, including while waiting for queries. When Tree mode prompts for a query, logging and progress start after the query is entered. The default log path uses the command‘s startup time: <target_dir>/log_export_data_yyyyMMddHHmmss.log. Override it with -l/--export_data_log <path>. The output directory is created automatically; an overridden log’s parent directory must exist.
The console is for people. In a terminal one progress line is redrawn in place, cutting the current measurement from the left to fit the terminal width. When output goes to a file, nohup or a container log, or when the progress line would not fit the terminal, each update is a line of its own:
Exporting 59% 121/204 3,535,481 rows 00:00:02 ETA 00:00:01 ...line19.s0 [59%] 121/204 measurements, 3,535,481 rows, elapsed 00:00:02, ETA 00:00:01, current root.factory.line19.s0
The shell launcher passes the terminal width to the tool as COLUMNS. The Windows launcher does not, so 80 columns are assumed there unless COLUMNS is set.
The console also shows the start, warnings and the log path, then ends with one summary line such as Export completely! Rows: 5, files: 1, elapsed: 00:00:00. Query and Table modes have no total, so they show rows, elapsed time and rows per second instead of a percentage and ETA. The launcher applies conf/logback-tool.xml when present, so library logging shows errors only.
The log is for programs: JSON Lines, one event object per line, with counts as JSON numbers. Each event starts with progress_id (the operating system process ID), timestamp (ISO 8601 with offset), model (tree or table), status (RUNNING, SUCCESS, FAILED or CANCELLED), exported_data_rows, elapsed_time (HH:mm:ss) and output, the file being written (omitted until a task opens one), followed by event, level, phase and the event's details. Events include START, DISCOVERY, WARNING, TASK_START, TASK_COMPLETE, FILE_START, FILE_COMPLETE, PROGRESS and one FINISH. The log records source patterns, queries, output files, rows, completed files, elapsed time and errors; command credentials are excluded. For example, jq -c 'select(.event == "FINISH")' export/log_export_data_*.log prints each result.
Measurement-file mode adds exported_measurement_num, total_measurement_num, exported_percent and estimated_remaining between exported_data_rows and elapsed_time. A measurement completes only after all output files and its query handle close successfully. estimated_remaining uses the average measurement duration (HH:mm:ss). Measurements differ in size, so it stays UNKNOWN, and the console shows ETA --, until 5% of the measurements, and at least three, have completed. No extra COUNT(*) scan is made, and no -progress option is needed.
The command returns 0 on success and 1 for invalid arguments, input, queries, output or log errors; failures record FAILED. Ctrl+C or SIGTERM records CANCELLED. The export stops at the next row and closes its query and session; a server call that is still blocked after three seconds is closed. The process then exits with the signal's status (130 for Ctrl+C, 143 for SIGTERM), which FINISH reports as exit_code. Already generated files are retained for inspection, and incomplete files are not marked FILE_COMPLETE.