Checkpoint Lab — Regular Expressions, CSS/JQuery, XPath, JSONPath, and Boundary Extraction

1. Assumptions and hard ceilings

2. Start and preflight the fixture

mkdir -p results/checkpoint results/broken-xml results/repaired-xml results/overhead
python fixtures/extraction_fixture.py --log results/checkpoint/server-events.jsonl

curl --fail --silent http://127.0.0.1:8000/health
curl --fail --silent http://127.0.0.1:8000/json
curl --fail --silent http://127.0.0.1:8000/xml
curl --fail --silent http://127.0.0.1:8000/html
curl --fail --silent http://127.0.0.1:8000/text

3. Select the extractor for each response

Set Match Number 1 and default __NOT_FOUND__ for every required primary extraction.

4. Write predictions before execution

1. Each source sampler returns HTTP 200 and creates exactly one expected primary variable.
2. Every /verify request returns 200 only if the extractor selected the correct ID for its source.
3. Changing XML from /xml to /xml-v2 will break the deliberately absolute XPath /response/primary/id/text() but not //primary/id/text().
4. Extractor runs may slightly increase whole-engine wall time or generator CPU while source sampler p50/p95 stays close to the source-only baseline.

5. Bounded authoring proof

Run one thread × one loop in GUI mode with Debug Sampler. Verify these thread variables:

JSON_ID          = JSON-PRIMARY-100
XML_ID           = XML-PRIMARY-200
HTML_ID          = HTML-PRIMARY-300
TEXT_BOUNDARY_ID = TEXT-PRIMARY-400
TEXT_REGEX_ID    = TEXT-PRIMARY-400

Also test all-match candidate selectors and record their match counts/numbered variables. Remove/disable Debug Sampler and View Results Tree afterward.

6. Run the functional CLI profile

jmeter -n   -t plans/checkpoint-extraction.jmx   -l results/checkpoint/results.jtl   -j results/checkpoint/jmeter.log   -Jjmeter.save.saveservice.print_field_names=true   -Jjmeter.save.saveservice.thread_counts=true

python tools/analyze_extraction.py results/checkpoint/results.jtl
curl --fail --silent http://127.0.0.1:8000/stats

Expected: no failed verification samples. If a verification returns 400, inspect the extractor variable/sentinel and server event before changing the target.

7. Optional two-thread proof

Run at most 2 threads × 2 loops. Extracted variables remain thread-local, so two users can independently parse/use the same synthetic endpoint structure without sharing a mutable JMeter property.

The fixture IDs are static per format for teaching selector behavior, so thread isolation is proven by variable scope/independent executions rather than uniqueness of the returned ID. Chapter 09 covered unique per-thread issued tokens.

8. Deliberately break XPath after a structure change

Create plans/broken-xml.jmx from the normal plan and make only two changes:

1. XML source path becomes /xml-v2.
2. XPath2 query becomes brittle /response/primary/id/text().

Run one thread × one loop. Preserve:

• XML_ID=__NOT_FOUND__ evidence;
• failed extraction-success assertion;
• JTL and jmeter.log;
• server response showing the ID still exists under <data>.

9. Restore the selector contract

If the wrapper is non-contractual, restore //primary/id/text() and rerun into results/repaired-xml/. Expected: extraction and verify request return to success without target modification.

If exact hierarchy is contractual in your real system, do not loosen the XPath: instead classify the response shape as a target contract regression. The checkpoint requires you to state which interpretation applies.

10. Measure small-scale extractor overhead

For each source format, create a source-only baseline and a one-extractor variant, each 1 thread × 20 loops. Time the command externally and record generator CPU/memory:

# Example pair for JSON
/usr/bin/time -p jmeter -n   -t plans/overhead-json-baseline.jmx   -l results/overhead/json-baseline.jtl   -j results/overhead/json-baseline.log

/usr/bin/time -p jmeter -n   -t plans/overhead-json-extractor.jmx   -l results/overhead/json-extractor.jtl   -j results/overhead/json-extractor.log

python tools/compare_runs.py   results/overhead/json-baseline.jtl   results/overhead/json-extractor.jtl

On PowerShell, use Measure-Command around each jmeter.bat -n ... invocation. Repeat the method for XML, HTML, and Text if desired, always comparing each extractor with its own identical source baseline.

11. Interpret overhead carefully

Do not claim “JSONPath is faster than XPath2” from these different payloads. The valid comparison is within one format: same source/workload with and without its extractor. If differences are below noise, report that.

Sampler elapsed largely reflects the HTTP operation. Extraction is Post-Processor work, so use whole-run wall time and generator CPU/GC as primary overhead evidence.

12. Required evidence packet

13. Verification checklist

• All traffic is loopback:8000.
• JSON uses JSONPath; XML uses XPath2; HTML uses current CSS Selector; text uses Boundary plus regex comparison.
• Required extractors use explicit sentinel defaults.
• Downstream /verify independently rejects wrong IDs.
• Match-all mode is limited to bounded debug/candidate examples.
• Broken absolute XPath evidence is preserved before repair.
• JTL and matching jmeter.log exist for meaningful runs.
• Generator observation accompanies overhead claims.

14. Validity statement

15. Cleanup

1. Stop the Python fixture.
2. Keep normal/broken/repaired evidence until review is complete.
3. Disable debug listeners and synthetic variable dumps in future load plans.
4. Do not copy these synthetic defaults into real credential/session extraction.
5. No production target, credential, recorder certificate, remote engine, container, database, paid service, plugin, or system-wide JVM/OS setting was changed.

16. What Chapter 10 adds to the operating model

The performance-testing operating model now has an extraction contract: expected response representation, selector language, scope, match count, default/failure behavior, produced thread variables, downstream consumers, privacy handling, and generator cost are reviewable inputs.

Chapter 11 builds on this by controlling the other side of dynamic data: CSV Data Set Config, parameterization, unique test data, exhaustion/recycling behavior, and safe multi-user data strategy.

Knowledge check

Official references and version notes

• Component Reference — current Regular Expression, CSS Selector, XPath2/XPath, JSON JMESPath, JSON, and Boundary Extractor semantics.
• Elements of a Test Plan — Post-Processor scope/execution and thread-local JMeter variables.
• Regular Expressions — JMeter regular-expression guidance and extractor examples.
• Best Practices — non-GUI load execution, lean listeners, generator validity, and scripting guidance.
• Apache JMeter downloads — current production release and Java requirement.