Bugzilla

The main change here is that we currently allow only 100% accurate quota management. We want to change it so we will allow "inaccurate" usage tracking for the time the temporary storage is being initialized. So quota clients will be able to use storage even if the initialization is not finished yet. When the initialization finishes and we see that more data has been written than it would be allowed with synchronous initialization, we will evict some origins. After that we will have 100% accurate quota management/usage tracking again.
Once we do necessary changes for initializing origins asynchronously, then they naturally won't break entire temporary storage initialization. The quota management will stay in "not 100% accurate" mode since we couldn't get exact usage for broken origins, The broken origins will stay uninitialized with some files on disk and they won't be included in overall usage calculations. The fact that we leave some extra files on disk which are not included in the usage calculations shouldn't be a big problem. We already had to make an exception for LSNG which tracks only logical size of the database. So the total physical size of all files doesn't have to match the usage we internally use for quota checks. We only allow to use 50% of free disk space anyway, so there should be a lot of space in reserve.

The main change here is that we currently allow only 100% accurate quota management. We want to change it so we will allow "inaccurate" usage tracking for the time the temporary storage is being initialized. So quota clients will be able to use storage even if the initialization is not finished yet. When the initialization finishes and we see that more data has been written than it would be allowed with synchronous initialization, we will evict some origins. After that we will have 100% accurate quota management/usage tracking again.
Once we do necessary changes for initializing origins asynchronously, then they naturally won't be able to break entire temporary storage initialization. The quota management will stay in "not 100% accurate" mode since we couldn't get exact usage for broken origins, The broken origins will stay uninitialized with some files on disk and they won't be included in overall usage calculations. The fact that we leave some extra files on disk which are not included in the usage calculations shouldn't be a big problem. We already had to make an exception for LSNG which tracks only logical size of the database. So the total physical size of all files doesn't have to match the usage we internally use for quota checks. We only allow to use 50% of free disk space anyway, so there should be a lot of space in reserve.

The main point here is that we currently allow only 100% accurate quota management. We want to change it so we will allow "inaccurate" usage tracking for the time the temporary storage is being initialized. So quota clients will be able to use storage even if the initialization is not finished yet. When the initialization finishes and we see that more data has been written than it would be allowed with synchronous initialization, we will evict some origins. After that we will have 100% accurate quota management/usage tracking again.
Once we do necessary changes for initializing origins asynchronously, then they naturally won't be able to break entire temporary storage initialization. The quota management will stay in "not 100% accurate" mode since we couldn't get exact usage for broken origins, The broken origins will stay uninitialized with some files on disk and they won't be included in overall usage calculations. The fact that we leave some extra files on disk which are not included in the usage calculations shouldn't be a big problem. We already had to make an exception for LSNG which tracks only logical size of the database. So the total physical size of all files doesn't have to match the usage we internally use for quota checks. We only allow to use 50% of free disk space anyway, so there should be a lot of space in reserve.