Crash server - Blacklist item during sync restart #1193
Conversation
…hub.com:Infomaniak/desktop-kDrive into Crash-server-blacklist-item-during-sync-restart
|
|
||
| class SYNCENGINE_EXPORT SyncPal : public std::enable_shared_from_this<SyncPal> { | ||
| public: | ||
| static std::recursive_mutex updateTreesMutex; |
There was a problem hiding this comment.
I am not comfortable with the idea of a mutex being accessible anywhere in the code. This seems to have high risk of deadlock and/or impacting performance.
Moreover, it seems to me that only the running module should be accessing and modifying the update trees. Therefor, no concurrent access should occur. Maybe the algorithm is to be refactored somewhere instead?
There was a problem hiding this comment.
The LocalFileSystemObserverWorker updates the update trees and runs in parallel with the other modules:
LocalFileSystemObserverWorker::sendAccessDeniedError => SyncPal::handleAccessDeniedItem => UpdateTree::deleteNode
|
During a sync restart after a DataError, the localFileSystemObserverWorker continues to run and in case of AccessDenied, the handleAccessDeniedItem method is called which accesses the updateTree while it is destroyed/recreated by the sync restart.