What breaks in document AI

Upgrades, backup, restore

22 of 22 quotes

• “How do I go back to the previous model?”

• “is now permanently frozen”

• “Upgrading from v0.25.5 to v0.26.4 via `mysql_migration.py` renders all pre-configured models unusable.”

• “5-user SQLite db migrated from v0.4.7 → v0.9.6, hard-crashed on the second row with the orphaned `info_json` column blocking restart”

• “I have 3x16GB cards and, normally, ollama would allow 48GB to be used. But now, it's like 32GB would be used.”

• Hit this one too? We are building for it.

• “During table recognition/parsing, multiple rows are now incorrectly merged into a single line.”

• “since 30.x there is trouble to run bigger LLMs, which was no issue with 0.24”

• “My physical machine crashed (blue screen), which forced my RAGFlow virtual machine to shut down. When I started it up again, my MySQL data volume got corrupted. This situation has occurred twice.”

• “Failure to migrate from 0.8.12 to 0.9.5”

• “The PR adds idempotency (existence checks) to the 6 affected migration files mentioned in this issue.”

• “Our setup only broke on c1d2e3f4a5b6”

• “same get this upgrading from 0.8.10 to 0.9.2”

• “After upgrading to version 0.25.0, all metadata has been lost. The large language models that were previously connected through LM Studio have all become invalid.”

• “I met the same issue when upgrade from 0.23.x to 0.24.0 and had to regenerate ALL my metadatas. When I upgrade from 0.24.0 to 0.25.0 today, all my metadatas lost AGAIN... Now I rollbacked to 0.24.0.”

• “Loading and login issue after upgrading to 0.9.0 and 0.9.1”

• “After upgrading to v0.24.0, I noticed that metadata is now read from Elasticsearch instead of MySQL. As a result, all my existing metadata can no longer be displayed properly.”

• “Database migration chat → chat_message OOM-kills on large datasets (PostgreSQL/AlloyDB, ~75 GB)”

• “There is no built-in, reproducible way to “snapshot” the current model list and restore it elsewhere.”

• “The previous documents and users are all gone.”

• “Same issue for me. most of admin pages doesn't load.”

• “After upgrading from Ollama version 0.12.3 to 0.12.4~0.13.2, I observed a noticeable degradation in the first token generation speed when running the `qwen3:32b` model.”

• “然后想把我的知识库数据迁移到另一个windows台式机上，当我按照步骤操作时，在使用.sh文件恢复时总失败”

• Hit this one too? We are building for it.