marcellourbani/client/media/skills/abap-performance-hana

ABAP Performance — S/4HANA / HANA Database

These rules apply to SAP S/4HANA systems or any ABAP system running on HANA DB.

Before using this skill: Call the SAP system info tool. If the system is ECC on a traditional DB, use the abap-performance-ecc skill instead.

Core philosophy on HANA: Push data-intensive operations to the database. HANA is a columnar in-memory DB optimized for set-based operations, aggregations, and complex SQL. Let it do the heavy lifting. Keep ABAP for business logic, authorization, and exception handling.

---

Code Pushdown — The #1 Rule

Move data-intensive operations to the database layer.

Push down to HANA:

• Aggregations (SUM, COUNT, AVG, MIN, MAX)
• Filtering (WHERE clauses — the more selective, the better)
• Sorting (ORDER BY)
• JOINs (HANA handles complex multi-table JOINs efficiently)
• String operations and arithmetic in SQL
• CASE expressions / conditional logic on data
• Grouping and HAVING
• Window functions (OVER/PARTITION BY)
• UNION / INTERSECT / EXCEPT

Keep in ABAP:

• Complex business logic with many branches
• Authority checks, messages, exceptions
• Small dataset processing where pushdown overhead exceeds benefit
• Operations requiring ABAP runtime features (RFC calls, file I/O, etc.)

Avoid:

• Reading all rows to ABAP and filtering/aggregating in loops
• Using internal tables as intermediate storage for what SQL can do in one statement
• Multiple sequential SELECTs that could be a single JOIN

---

Database Access

SELECT Patterns

• Select only the fields you need. Never SELECT * in production.

  SELECT matnr, maktx FROM mara INTO TABLE @DATA(itab).
  

• Always use a WHERE clause. Always use @ escaped host variables.

• Use JOINs aggressively — HANA handles complex JOINs very well, even 5+ tables.

  SELECT m~matnr, t~maktx, p~werks, p~ekgrp
    FROM mara AS m
    INNER JOIN makt AS t ON t~matnr = m~matnr AND t~spras = @sy-langu
    INNER JOIN marc AS p ON p~matnr = m~matnr
    LEFT OUTER JOIN mvke AS s ON s~matnr = m~matnr
    WHERE m~mtart = @material_type
    INTO TABLE @DATA(materials).
  

• Use aggregate functions and GROUP BY — let HANA calculate:

  SELECT werks, SUM( labst ) AS total_stock, COUNT(*) AS item_count
    FROM mard
    WHERE matnr = @matnr
    GROUP BY werks
    INTO TABLE @DATA(stock_by_plant).
  

• Use CASE expressions to push conditional logic to the DB:

  SELECT matnr,
         CASE mtart
           WHEN 'FERT' THEN 'Finished'
           WHEN 'ROH' THEN 'Raw Material'
           ELSE 'Other'
         END AS type_text
    FROM mara
    INTO TABLE @DATA(materials).
  

• Use string functions in SQL:

  SELECT matnr, CONCAT( matnr, CONCAT( ' - ', maktx ) ) AS display_text
    FROM mara
    INNER JOIN makt ON makt~matnr = mara~matnr AND makt~spras = @sy-langu
    INTO TABLE @DATA(display_data).
  

• Use FOR ALL ENTRIES when JOINs aren't possible. Always check driver table is not empty.

• Use UP TO n ROWS when you only need limited results.

• Use subqueries where they simplify logic:

  SELECT matnr, maktx FROM mara
    WHERE matnr IN ( SELECT matnr FROM marc WHERE werks = @plant )
    INTO TABLE @DATA(plant_materials).
  

CDS Views

• Prefer CDS views for complex data models. They are the primary code pushdown mechanism on HANA.
• CDS views are reusable, testable, and automatically optimized by HANA.
• Use CDS for: complex joins, calculated fields, aggregations, associations, access control.
• Consume CDS views in ABAP via SELECT FROM zcds_view.

AMDP (ABAP-Managed Database Procedures)

• Use AMDP for very complex calculations that must run entirely on HANA.
• AMDP gives you access to full SQLScript (HANA's procedural SQL language).
• Use when: complex multi-step transformations, heavy string processing, graph operations, or when CDS is insufficient.
• AMDP is NOT portable to other DBs — use only when you're certain the system stays on HANA.

Avoiding Redundant DB Access

• Never SELECT the same data twice. Read once, reuse.

• Use READ TABLE on internal table buffer instead of SELECT SINGLE in a loop:

  SELECT matnr, maktx FROM makt WHERE spras = @sy-langu INTO TABLE @DATA(texts).
  " later:
  READ TABLE texts WITH KEY matnr = current_matnr INTO DATA(text_line).
  

• On HANA, even redundant DB access is faster than on traditional DBs — but it's still wasteful and adds network overhead.

Table Buffering

Buffering matters less on HANA than ECC because HANA is in-memory. But it still helps for:

• Reducing network round-trips between app server and DB server

• Avoiding query parsing overhead for tiny lookups

• Use SELECT SINGLE on buffered tables — reads from buffer. UP TO 1 ROWS bypasses buffer.

  " good — uses buffer
  SELECT SINGLE * FROM t001 WHERE bukrs = @bukrs INTO @DATA(company).
  " bad — bypasses buffer
  SELECT * FROM t001 UP TO 1 ROWS WHERE bukrs = @bukrs INTO @DATA(company).
  

• JOINs, aggregates, GROUP BY, ORDER BY, subqueries bypass the buffer.

---

Internal Tables

Table Type Selection

Same rules as any ABAP system — this is ABAP runtime, not DB:

• HASHED: O(1) lookup. Large tables, unique key, read-heavy, filled once.
• SORTED: O(log n) lookup. Non-unique key, range access, incremental fill.
• STANDARD: O(n) unless sorted + binary search. Small tables or sequential.

" good — O(1) lookup
DATA materials TYPE HASHED TABLE OF mara WITH UNIQUE KEY matnr.
READ TABLE materials WITH TABLE KEY matnr = input INTO DATA(mat).

Loop Optimization

• Use ASSIGNING FIELD-SYMBOL(<fs>) for fastest loop processing.
• Use WHERE on LOOP — especially on SORTED tables.
• Avoid nested loops O(n*m). Use HASHED lookup for inner data.
• Use FILTER for extracting subsets from SORTED/HASHED tables:

  DATA(subset) = FILTER #( sorted_table WHERE status = 'A' ).
  

Bulk Operations

• INSERT lines_of for bulk inserts.
• VALUE #( FOR ... ) and REDUCE for functional transformations.
• CORRESPONDING #( ) for structure mapping.

HANA-Specific: Consider Pushing to SQL

Before writing a complex ABAP loop with aggregation, filtering, or transformation — ask: can this be a SQL statement instead?

" ABAP way (acceptable for small data)
LOOP AT sales ASSIGNING FIELD-SYMBOL(<s>).
  AT NEW kunnr.
    total = 0.
  ENDAT.
  total += <s>-netwr.
  AT END OF kunnr.
    APPEND VALUE #( kunnr = <s>-kunnr total = total ) TO totals.
  ENDAT.
ENDLOOP.

" HANA way (better for large data)
SELECT kunnr, SUM( netwr ) AS total
  FROM vbak
  WHERE erdat >= @from_date
  GROUP BY kunnr
  INTO TABLE @DATA(totals).

---

String Operations

• Use string templates | | instead of CONCATENATE.

• Avoid repeated string concatenation in loops — build a string table:

  DATA lines TYPE string_table.
  LOOP AT data INTO DATA(d).
    APPEND |{ d-field1 };{ d-field2 }| TO lines.
  ENDLOOP.
  DATA(csv) = concat_lines_of( table = lines sep = cl_abap_char_utilities=>cr_lf ).
  

• HANA-specific: For heavy string assembly from DB data, consider doing it in SQL with CONCAT or STRING_AGG (via CDS/AMDP).

---

Authorization Checks

• Check authority before expensive data retrieval:

  AUTHORITY-CHECK OBJECT 'M_MATE_WRK' ID 'WERKS' FIELD plant.
  IF sy-subrc <> 0. RAISE EXCEPTION NEW zcx_no_auth( ). ENDIF.
  SELECT ... " now fetch
  

• On S/4HANA, consider CDS access control (DCL) for row-level authorization built into the data model.

---

ALV / UI Performance

• Pass data by reference.
• Use CL_SALV_TABLE for read-only display.
• For very large result sets, consider pagination.
• On S/4HANA: consider Fiori/RAP for UI instead of classical ALV.

---

Parallel Processing

• aRFC for independent parallel tasks.
• SPTA framework for parallelized mass processing.
• Background jobs for very long tasks.
• HANA-specific: Before parallelizing in ABAP, check if the work can be pushed to HANA — a single efficient SQL may outperform parallel ABAP tasks.

---

HANA Anti-Patterns

| Anti-Pattern | Fix | |---|---| | SELECT * | Select only needed fields | | SELECT in a LOOP | JOINs (HANA handles complex JOINs well) | | Aggregating in ABAP loops | SUM/COUNT/AVG in SQL with GROUP BY | | Filtering in ABAP what SQL can filter | Push WHERE to SQL | | Multiple SELECTs that could be one JOIN | Combine into single JOIN statement | | Nested LOOPs on STANDARD tables | HASHED lookup for inner data | | Complex ABAP transformations on large data | CDS view or AMDP | | String concat in loops with && | Build string table, or push to SQL | | Ignoring CDS views | Use CDS for reusable data models | | UP TO 1 ROWS on buffered table | SELECT SINGLE to use buffer | | Authority check after data retrieval | Check before SELECT, or use CDS DCL | | Writing ABAP for what SQL can express | Push to database — always ask "can SQL do this?" |