From the lab / Oracle Data Pump

Recover an Oracle 11g dump, including BLOB images

Four tables out as CSV. The coffee mug comes out as a JPEG.

In the deleted-rows article, the coffee mug was just a name in a table. This time I gave it a photo, stored that photo in a BLOB, and put the whole desk kit into an Oracle 11g Data Pump dump.

I wanted the rows as CSV and the pictures as ordinary files, without importing the dump into another Oracle database.

1. Four tables from 11g.

I created a small schema, BLOG_DESK11, on Oracle 11.2.0.4 and exported its four tables with expdp. The dump was made for this test on an actual 11g instance.

The test schema — 23 rows
TableRowsWhat is in it
DESK_KIT15Item names, quantities, prices, dates and Unicode labels
ITEM_NOTES3CLOB and NCLOB text, a line break, quotes and NULLs
ITEM_PHOTOS2Two small desk-item images, each stored as a JPEG BLOB
TYPE_SAMPLES3CHAR, NUMBER, binary floats, RAW, timestamps, a time-zone offset and both interval types

The dump uses metadata-only compression. Extraction runs on a separate machine, reading the copied .dmp file. It does not connect to the source database.

2. Let the wizard do the export.

I start ./reclaimdb dmp-extract --wizard, open the dump and choose Pick by schema → BLOG_DESK11. Then CSV, a comma delimiter, RFC 4180 and a header row. At the destination step I keep side-files for LOBs larger than 4 KiB. That keeps the pictures out of the CSV cells.

Actual wizard session — schema selection, CSV options, BLOB side-files and export

Press play to replay the real inputs and output. Playback runs at 0.7× speed; long idle waits are shortened. On a narrow screen, scroll the terminal sideways.

The review screen shows all four tables and the equivalent command before I start. The run exports 15 + 3 + 2 + 3 rows.

3. Open the BLOBs as ordinary files.

The photo CSV contains file references:

ITEM_PHOTOS — actual CSV
ID,ITEM_ID,FILE_NAME,MIME_TYPE,PHOTO
1,6,desk-items.jpg,image/jpeg,ITEM_PHOTOS/0_PHOTO.jpg
2,1,notebook.jpg,image/jpeg,ITEM_PHOTOS/1_PHOTO.jpg

Those paths are relative to recovered/lobs/. So the first image is recovered/lobs/ITEM_PHOTOS/0_PHOTO.jpg. No hex conversion or Oracle tool needed: it is a JPEG you can open.

Click either image to open the JPEG recovered from the BLOB. I used generated test pictures, reduced to 480 × 320 pixels before loading them into Oracle.

All 23 rows recovered. The exported values matched the original Oracle data. Both image files were identical to the pictures loaded into the database.
Wizard worklog — inputs and terminal output

Blue marks what I entered; ↵ means Enter. These inputs and the output below come from the recorded run.

Wizard inputsActual session
$ ./reclaimdb dmp-extract --wizard

Begin                       ↵ Enter
Dump file                   source/desk-kit-11g.dmp
List all four tables        v
Continue                    ↵ Enter
Select by schema            s
Schema name                 BLOG_DESK11
CSV format                  9
Comma delimiter             ↵ Enter
RFC 4180 dialect            r
Include header row          ↵ Enter
Output directory            recovered
Write files, not stdout     ↵ Enter
LOB side-files              s
If files exist: error       ↵ Enter
Keep advanced defaults      ↵ Enter
Start export after review   ↵ Enter
Save settings               ↵ Enter
Wizard outputActual session
$ ./reclaimdb dmp-extract --wizard

  reclaimDB  dmp-extract wizard  v5.0.0-alpha5
  --------------------------------------------------------------------------------------------------------------------
  Extract Oracle Data Pump files to CSV, JSON, Parquet and more.

  Detected:  TTY on  Colour off  Unicode off  Width 120  Locale C.UTF-8

Question:   Press Enter to begin.

source/desk-kit-11g.dmp

  Step 1 of 9 — Input file
  --------------------------------------------------------------------------------------------------------------------
  No .dmp files in the current directory.

Question:   Path to a .dmp file or directory:

  Reading header...
  DMP XML enrich scan        0.2 /     0.2 MB     0.0 MB/s  ETA --:--  rows=0

  DMP table-def scan         0.2 /     0.2 MB    30.2 MB/s  ETA 00:00  rows=4

+ 11.02.00.00.00, unknown platform, unknown charset (4096-byte blocks)
  Detecting features... + metadata-only compression (row data plaintext), not encrypted
  Scanning tables...    + 4 tables · 244.0 KB

  Step 3 of 9 — What's in the file
  --------------------------------------------------------------------------------------------------------------------
  Source:   source/desk-kit-11g.dmp
  Oracle:   11.02.00.00.00  (charset unknown)
  Schemas:  BLOG_DESK11    4 tables total

  Largest-by-columns:
    BLOG_DESK11.TYPE_SAMPLES                   12 cols · 1 parts
    BLOG_DESK11.DESK_KIT                        6 cols · 1 parts
    BLOG_DESK11.ITEM_PHOTOS                     5 cols · 1 parts
    BLOG_DESK11.ITEM_NOTES                      4 cols · 1 parts

  [Enter] continue    [v] list all    [b] back
  v

  All tables (4):
    BLOG_DESK11.ITEM_PHOTOS  (5 cols, 1 parts)
    BLOG_DESK11.ITEM_NOTES  (4 cols, 1 parts)
    BLOG_DESK11.DESK_KIT  (6 cols, 1 parts)
    BLOG_DESK11.TYPE_SAMPLES  (12 cols, 1 parts)

  [Enter] continue

  Step 4 of 9 — Which tables
  --------------------------------------------------------------------------------------------------------------------
Choice:     [a] All tables  (recommended for a full extract)
Choice:     [s] Pick by schema
Choice:     [t] Pick by table      (fuzzy finder; or: `t HR.ORDERS HR.X` inline)
Choice:     [x] All except prefix
Choice:     [i] List tables (don't extract)
  > s

  Schema name:
  BLOG_DESK11

  Step 5 of 9 — Output format
  --------------------------------------------------------------------------------------------------------------------
  Grouped by intended use. Recommended for your data: *

    Analytics (columnar, for data lakes and BI):
      [1]   Parquet         widely supported, 3–10x smaller than CSV
      [2]   Apache Arrow    IPC (DuckDB, Pandas, Polars)
      [3]   Apache ORC      Apache ORC (Hive, Presto)
      [4]   Delta Lake      Parquet + _delta_log/
      [5]   Apache Iceberg  Parquet + metadata/

    Databases (load into another DB):
      [6]   PostgreSQL      PostgreSQL binary COPY + DDL
      [7]   SQL             portable INSERTs, dialect-aware
      [8]   SQLite          one .sqlite per table
      [l]   SQL*Loader      Oracle reimport via sqlldr (.dat + .ctl)

    Interchange (human-friendly):
      [9] * CSV             default delimiter: comma
      [0]   JSON            array per table
      [.]   NDJSON          JSON Lines
      [x]   XML             element per row
      [e]   Excel / XLSX    Excel .xlsx
      [v]   Apache Avro     Object Container (Kafka, Hadoop)

  > [Enter] accepts recommended (CSV)
  9

  Step 5b — CSV options
  --------------------------------------------------------------------------------------------------------------------
  Delimiter [Enter]=, [t]=tab [|]=pipe [c]=custom:
  CSV dialect [r]=rfc4180 [e]=excel-de [Enter]=rfc4180 (recommended for ','): r

  Header row? [Enter]=yes [n]=no:
  Step 6 of 9 — Where to write
  --------------------------------------------------------------------------------------------------------------------
  Output directory (Enter keeps default):
    default: /home/mfg82/.local/share/reclaimDB/2026-11-10-extract
  path> recovered
  [s] Write to stdout instead? [Enter]=no:
  LOBs >4 KiB [Enter]=side-files [i]=inline-hex [t]=truncate: s
  If files exist: [Enter]=error [w]=overwrite [s]=skip [a]=append:

  Step 7 of 9 — Advanced (skippable)
  --------------------------------------------------------------------------------------------------------------------
  [Enter] skip (keep defaults)    [A] open advanced options

  Step 8 of 9 — Review
  --------------------------------------------------------------------------------------------------------------------
  Input:    source/desk-kit-11g.dmp
  Source:   11.02.00.00.00  (charset?)
  Tables:   4 detected in scan
  Mode:     pick by schema (BLOG_DESK11)
  Format:   CSV (delim=',', rfc4180)
  Output:   recovered
  Estimate: input 0.2 MB · approx 1 s at 100 MB/s

  Equivalent command:
    dmp-extract \
  --csv recovered \
  --schema BLOG_DESK11 \
  --csv-dialect rfc4180 \
  --csv-delimiter ',' \
  --lob-dir recovered/lobs \
  source/desk-kit-11g.dmp
  Project file will be saved to: ./extract-2026-11-10T18-17-20Z.json

  [Enter] start   [e] edit a step   [D] dry run   [s] save and exit   [b] back
  (Ctrl+E dumps equivalent CLI, Ctrl+S saves now, Ctrl+V verbose)
  > Progress:
  Extracting...
  --------------------------------------------------------------------------------------------------------------------

    ╭────┬╌╌╌╮                     _         _             ____   ____
    │████│╌╌╌┊   _ __   ___   ___ | |  __ _ (_) _ __ ___  |  _ \ | __ )
    │████│   ┊  | '__| / _ \ / __|| | / _` || || '_ ` _ \ | | | ||  _ \
    │████│╌╌╌┊  | |   |  __/| (__ | || (_| || || | | | | || |_| || |_) |
    ╰────┴╌╌╌╯  |_|    \___| \___||_| \__,_||_||_| |_| |_||____/ |____/
                       recover  ·  v4.19.0-rc1

                Oracle Data Pump Extractor · v4.19.0-rc1

  Memory governor: MemAvailable=33.7 GB, effective_total=43.7 GB, reserve=4.4 GB, budget=29.3 GB, DMP input=244.0 KB, largest piece=244.0 KB

  DMP XML enrich scan        0.2 /     0.2 MB     0.0 MB/s  ETA --:--  rows=0

  DMP table-def scan         0.2 /     0.2 MB    24.5 MB/s  ETA 00:00  rows=1

  Source:   source/desk-kit-11g.dmp
  Size:     244.0 KB
  Version:  3.1
  Platform:
  Oracle:   11.02.00.00.00
  Charset:  (unknown)
  Block:    4096 bytes
  Created:  2026-11-10 19:14:30

  Compression: METADATA_ONLY — metadata is compressed, row data is plaintext; extracting directly
  Metadata: 16.7 KB decompressed from 1 BASIC unit(s) for composite types

  DMP XML enrich scan        0.2 /     0.2 MB     0.0 MB/s  ETA --:--  rows=0

  DMP table-def scan         0.2 /     0.2 MB    24.2 MB/s  ETA 00:00  rows=1

  DMP table-def scan         0.1 /     0.1 MB    17.8 MB/s  ETA 00:00  rows=4

  DMP table-def scan         0.1 /     0.1 MB    11.6 MB/s  ETA 00:00  rows=3

  DMP table-def scan         0.1 /     0.1 MB    12.0 MB/s  ETA 00:00  rows=2

  DMP table-def scan         0.1 /     0.1 MB    12.4 MB/s  ETA 00:00  rows=1

  Selective read: 4 stream(s) resolved via master index (DUMP_POSITION); full scan skipped

  DMP XML enrich scan        0.2 /     0.2 MB     0.0 MB/s  ETA --:--  rows=0

  DMP table-def scan         0.2 /     0.2 MB    23.6 MB/s  ETA 00:00  rows=1

  4 table(s) (filtered):
    BLOG_DESK11.ITEM_PHOTOS  5 cols
    BLOG_DESK11.ITEM_NOTES  4 cols
    BLOG_DESK11.DESK_KIT  6 cols
    BLOG_DESK11.TYPE_SAMPLES  12 cols

  ✓ BLOG_DESK11.ITEM_PHOTOS                   2 rows
  ✓ BLOG_DESK11.ITEM_NOTES                    3 rows
  ✓ BLOG_DESK11.DESK_KIT                     15 rows
  ✓ BLOG_DESK11.TYPE_SAMPLES                  3 rows

  ✓  23 rows exported to recovered  (0.0s, 72 MB/s)

  Note: no Oracle DDL sidecar for BLOG_DESK11.DESK_KIT: the table's column list was not found.
  Note: no Oracle DDL sidecar for BLOG_DESK11.TYPE_SAMPLES: the table's column list was not found.

  DMP extract rows           0.2 /     0.2 MB    52.7 MB/s  ETA 00:00  rows=23

  Done
  --------------------------------------------------------------------------------------------------------------------
  + Extraction complete.

  Next steps:
    · Inspect:  ls -lh recovered
    · Verify your restore: enable --audit on this workflow, then run
        reclaimdb verify-manifest ./manifest-<run-id>.json

  Save settings for replay? [Enter]=yes to ./extract-2026-11-10T18-17-20Z.json  [n]=no:
Project screen log: ./logs/run-2026-11-10T18:16:03Z-1111804-284404700-0.log
  Project saved: ./extract-2026-11-10T18-17-20Z.json

  ✓ Extraction complete
    Source: source/desk-kit-11g.dmp
    Format: CSV
    Output: recovered
    Inspect: ls -lh recovered
    Verify your restore: enable --audit, then run reclaimdb verify-manifest ./manifest-<run-id>.json

For the broader workflow, see Oracle dump extraction, or request a reclaimDB evaluation.

Oracle 11g · Data Pump · CSV · BLOB images

← All posts