Mercurial > hg > orthanc-databases
view TODO @ 582:8296c6a0238e find-refactoring
ODBC: fixes limit
author | Alain Mazy <am@orthanc.team> |
---|---|
date | Mon, 21 Oct 2024 17:57:52 +0200 |
parents | 7057d9db8d9a |
children |
line wrap: on
line source
============================== == Orthanc database plugins == ============================== ----------------------- Common - Database index ----------------------- * Performance of joins in LookupResources: Create cached statement for LookupResources, that are grouped to search up to, say, 10 tags, instead of recompiling for each request * Do not log "DatabaseCannotSerialize" errors in the plugin but only in Orthanc after all retries have been made. * Try to avoid the use of temporary tables: https://discourse.orthanc-server.org/t/image-insert-are-too-slow-databse-performance-too-poor/3820 * Implement "large queries" for: - updating all metadata of a resource at once - update all maindicomtags of 4 resource levels at once --------------------- Common - Storage area --------------------- * Implement the "StoreDicom" option from the default filesystem storage area to run MySQL/PostgreSQL storage in index-only mode: https://orthanc.uclouvain.be/book/contributing.html * Performance: Add a pool of connections to PostgreSQL/MySQL, as already done in the database index plugins ---------- PostgreSQL ---------- * Check if we can force the schema that is used. By default, Orthanc is using the 'public' schema but, after a wrong command, we have seen a DB where there was a 'AttachedFiles' table in the public schema and another one in a 'MyPacs' schema and Orthanc was actually using the 'MyPacs.AttachedFiles' table !!! Orthanc was then seeing only the most recent attached files !!! * Have a separate "thread" to UpdateStatistics to avoid large computations ? * Seems Orthanc might deadlock when there are plenty of conflicting transactions: https://groups.google.com/g/orthanc-users/c/xQelEcKqL9U/m/HsvxwlkvAQAJ https://groups.google.com/g/orthanc-users/c/1bkClfZ0KBA/m/s4AlwVh3CQAJ * use ReadCommited mode by default * use 40 index connection count by default (not 50 because, by default, PG has 100 connections and we want at least a few connections to remain available for sys admins if e.g, 2 Orthanc instances are running on the same server) ----- MySQL ----- * Store revisions for metadata and attachments in MySQL (this is already implemented in PostgreSQL) * MySQL performance => implement GlobalProperty_GetTotalSizeIsFast: https://groups.google.com/d/msg/orthanc-users/kSR4a110zDo/D7e4ITR8BwAJ https://discourse.orthanc-server.org/t/when-running-housekeeping-mysqld-exe-searching-and-dicom-transfer-is-slow/4921/9 * Add index to speed up wildcard search, as already done in PostgreSQL: - https://dev.mysql.com/doc/refman/5.5/en/fulltext-search.html + ALTER TABLE table ADD FULLTEXT index_name(column1); - https://dev.mysql.com/doc/refman/8.0/en/index-btree-hash.html ---- ODBC ---- * Implement labels