LLM_EXTRACT_TEXT implementation

[charleschile]

What type of PR is this?

• API-change
• BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

issue #18664

What this PR does / why we need it:

As part of our document LLM support, we are introducing the LLM_EXTRACT_TEXT function. This function extracts text from PDF files and writes the extracted text to a specified text file, extractor type can be specified by the third argument.

Usage: llm_extract_text(<input PDF datalink>, <output text file datalink>, <extractor type string>);

Return Value: A boolean indicating whether the extraction and writing process was successful.

Note:

• Both the input and output paths must be absolute paths.

Example SQL:

select llm_extract_text(cast('file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.pdf?offset=0&size=4' as datalink), cast('file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.txt' as datalink), "pdf");

Example return:

mysql> select llm_extract_text(cast('file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.pdf?offset=0&size=4' as datalink), cast('file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.txt' as datalink), "pdf");
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| llm_extract_text(cast(file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.pdf?offset=0&size=4 as datalink), cast(file:///Users/charles/Desktop/codes/matrixone/matrixone/test/distributed/resources/llm_test/extract_text/example.txt as datalink), pdf)   |
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| true                                                                                                                                                                                                                                                                                                                    |
+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.10 sec)

[qodo-merge-pro]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Reviewer Guide 🔍

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 No relevant tests

🔒 Security concerns Potential security vulnerability: The removal of file type validation in the ParseDatalink function (pkg/container/types/datalink.go) could potentially allow processing of unexpected file types, which may lead to security risks such as arbitrary file read or write operations. This change should be carefully reviewed to ensure it doesn't introduce vulnerabilities in the system.

⚡ Key issues to review Potential Security Risk Removal of file type validation in ParseDatalink function may lead to security vulnerabilities by allowing processing of unexpected file types. Error Handling The function does not properly handle errors in the PDF reading process. It should return an error instead of a boolean. Resource Management The PDF file is not properly closed in case of an error. Consider using defer immediately after successful open.

[qodo-merge-pro]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Code Suggestions ✨

Category	Suggestion	Score
Error handling	✅ Improve error handling by returning an error instead of a boolean Suggestion Impact: The function extractTextFromPdfAndWriteToFile was modified to return an error along with a boolean, improving error handling by providing more context about failures. code diff: -func extractTextFromPdfAndWriteToFile(pdfPath string, txtPath string, proc *process.Process) bool { +func extractTextFromPdfAndWriteToFile(pdfPath string, txtPath string, proc *process.Process) (bool, error) { // read PDF to string content, err := readPdfToString(pdfPath) if err != nil { - return false + return false, moerr.NewInvalidInputNoCtxf("Invalid PDF input.") } // file service and write file ctx := context.TODO() fs, readPath, err := fileservice.GetForETL(ctx, proc.Base.FileService, txtPath) - // NOTE: Write only works when a file does not exist + // delete the file if txt file exist because Write() only works when a file does not exist + _, err = fs.StatFile(ctx, readPath) + if err == nil { + err1 := fs.Delete(ctx, readPath) + if err1 != nil { + return false, moerr.NewInvalidInputNoCtxf("Cannot remove file %s", readPath) + } + } + _, err = fileservice.DoWithRetry( "BackupWrite", func() (int, error) { @@ -96,9 +125,9 @@ fileservice.IsRetryableError, ) if err != nil { - return false + return false, err } - return true + return true, nil Consider using a more robust error handling mechanism. Instead of returning a boolean, return an error that can provide more context about what went wrong during the extraction process. pkg/sql/plan/function/func_llm.go [69-74] -func extractTextFromPdfAndWriteToFile(pdfPath string, txtPath string, proc *process.Process) bool { +func extractTextFromPdfAndWriteToFile(pdfPath string, txtPath string, proc *process.Process) error { // read PDF to string content, err := readPdfToString(pdfPath) if err != nil { - return false + return fmt.Errorf("failed to read PDF: %w", err) } Apply this suggestion Suggestion importance[1-10]: 9 Why: The suggestion to return an error instead of a boolean provides more context about failures, improving error handling and debugging capabilities significantly.	9
Enhancement	Add context parameter for better control over function execution Consider adding a context parameter to the extractTextFromPdfAndWriteToFile function to allow for cancellation and timeout handling. pkg/sql/plan/function/func_llm.go [69] -func extractTextFromPdfAndWriteToFile(pdfPath string, txtPath string, proc *process.Process) bool { +func extractTextFromPdfAndWriteToFile(ctx context.Context, pdfPath string, txtPath string, proc *process.Process) bool { Apply this suggestion Suggestion importance[1-10]: 8 Why: Adding a context parameter allows for better control over function execution, enabling cancellation and timeout handling, which is a valuable enhancement for long-running operations.	8
	Enhance PDF parsing robustness and error handling Consider implementing a more robust PDF parsing logic that can handle different PDF structures and potential parsing errors. pkg/sql/plan/function/func_llm.go [112-121] func readPdfToString(path string) (string, error) { f, r, err := pdf.Open(path) if err != nil { - return "", err + return "", fmt.Errorf("failed to open PDF: %w", err) } - defer func() { - if f != nil { - f.Close() - } - }() + defer f.Close() + + if r.IsEncrypted() { + return "", fmt.Errorf("encrypted PDFs are not supported") + } Apply this suggestion Suggestion importance[1-10]: 8 Why: Improving PDF parsing logic to handle different structures and potential errors enhances the robustness and reliability of the function, making it more resilient to various input conditions.	8
Performance	Implement concurrent processing of PDFs for improved performance Consider using a buffered channel or a worker pool to process multiple PDFs concurrently, which could improve performance for large datasets. pkg/sql/plan/function/func_llm.go [22-29] +results := make(chan bool, rowCount) for i := uint64(0); i < rowCount; i++ { - inputBytes, nullInput := input.GetStrValue(i) - if nullInput { - if err := rs.AppendMustNullForBytesResult(); err != nil { - return err + go func(index uint64) { + inputBytes, nullInput := input.GetStrValue(index) + if nullInput { + results <- false + return } - continue + // Process PDF and send result + results <- extractTextFromPdfAndWriteToFile(...) + }(i) +} + +for i := uint64(0); i < rowCount; i++ { + success := <-results + if err := rs.Append(success, false); err != nil { + return err } +} Apply this suggestion Suggestion importance[1-10]: 7 Why: The suggestion to use concurrent processing can significantly improve performance for large datasets, although it introduces complexity and requires careful handling of concurrency issues.	7

[fengttt]

How do you know you are extracting a pdf file? How about add a const "file type" or "extractor type" arg to the function?

Extractor type arg might be better, if we happen to have many different kind of extractors for same file type (or a extractor for many file types).

[fengttt]

Please add a test.

[charleschile]

How do you know you are extracting a pdf file? How about add a const "file type" or "extractor type" arg to the function?

Extractor type arg might be better, if we happen to have many different kind of extractors for same file type (or a extractor for many file types).

Extractor type arg added

[charleschile]

Please add a test.

Unit and BVT tests have been added.

[fengttt]

Please add result verification in the bvt test.

[mergify]

⚠️ The sha of the head commit of this PR conflicts with #19725. Mergify cannot evaluate rules on this PR. ⚠️