Music scanning error

[fraenhawk 58]

I was getting sick of filling in all the musicbrainz artist/album/group strings manually and yet still having albums show up with no album artist so I decided to try out Picard and see if having everything embedded in the tags would help (although I would think the mediabrowser server should trump any file tags). So I started small and just had it identify 2 albums and even though Picard embedded all the tags, the metadata manager was only showing the album group after a rescan (no artist or album ID filled in). Is that normal?

 

So I decided to restart the server since every so often I've seen some odd caching behavior in MBS. Well, after restart I can't complete a scan and it always fails when starting to analyze music. Did something happen from Picard??

2015-01-23 21:02:24.1186 Error - App: Error
	*** Error Report ***
	Version: 3.0.5490.2
	Command line: C:\Users\Administrator\AppData\Roaming\MediaBrowser-Server\System\MediaBrowser.ServerApplication.exe
	Operating system: Microsoft Windows NT 6.1.7601 Service Pack 1
	Processor count: 2
	64-Bit OS: True
	64-Bit Process: False
	Program data path: C:\Users\Administrator\AppData\Roaming\MediaBrowser-Server
	Application Path: C:\Users\Administrator\AppData\Roaming\MediaBrowser-Server\System\MediaBrowser.ServerApplication.exe
	Type definitions should start with a '{', expecting serialized type 'Audio', got string starting with: "{\"FormatName\":"mp3",\"Size\":7591065,\"Containe
	System.Runtime.Serialization.SerializationException
	   at ServiceStack.Text.Common.DeserializeTypeRefJson.StringToType(TypeConfig typeConfig, String strType, EmptyCtorDelegate ctorFn, Dictionary`2 typeAccessorMap)
	   at ServiceStack.Text.Common.DeserializeType`1.<>c__DisplayClass3.<GetParseMethod>b__1(String value)
	   at ServiceStack.Text.JsonSerializer.DeserializeFromString(String value, Type type)
	   at ServiceStack.Text.JsonSerializer.DeserializeFromStream(Type type, Stream stream)
	   at MediaBrowser.Common.Implementations.Serialization.JsonSerializer.DeserializeFromStream(Stream stream, Type type)
	   at MediaBrowser.Server.Implementations.Persistence.SqliteItemRepository.GetItem(IDataReader reader)
	   at MediaBrowser.Server.Implementations.Persistence.SqliteItemRepository.<GetChildrenItems>d__11.MoveNext()
	   at System.Linq.Enumerable.WhereSelectEnumerableIterator`2.MoveNext()
	   at System.Linq.Enumerable.WhereEnumerableIterator`1.MoveNext()
	   at System.Collections.Generic.List`1..ctor(IEnumerable`1 collection)
	   at System.Linq.Enumerable.ToList[TSource](IEnumerable`1 source)
	   at MediaBrowser.Controller.Entities.Folder.get_ActualChildren()
	   at MediaBrowser.Controller.Entities.Folder.GetActualChildrenDictionary()
	   at MediaBrowser.Controller.Entities.Folder.<ValidateChildrenInternal>d__a6.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateSubFolders>d__d1.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.ValidateEnd(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateChildrenInternal>d__a6.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateSubFolders>d__d1.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.ValidateEnd(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateChildrenInternal>d__a6.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateSubFolders>d__d1.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.ValidateEnd(Task task)
	   at MediaBrowser.Controller.Entities.Folder.<ValidateChildrenInternal>d__a6.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.ValidateEnd(Task task)
	   at MediaBrowser.Server.Implementations.Library.LibraryManager.<PerformLibraryValidation>d__57.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Server.Implementations.Library.LibraryManager.<ValidateMediaLibraryInternal>d__50.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Common.Implementations.ScheduledTasks.ScheduledTaskWorker.<>c__DisplayClassf.<<ExecuteTask>b__e>d__11.MoveNext()
	--- End of stack trace from previous location where exception was thrown ---
	   at System.Runtime.CompilerServices.TaskAwaiter.ThrowForNonSuccess(Task task)
	   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
	   at MediaBrowser.Common.Implementations.ScheduledTasks.ScheduledTaskWorker.<Execute>d__6.MoveNext()
	
2015-01-23 21:02:24.1186 Info - App: Scan media library Failed after 0 minute(s) and 10 seconds

[fraenhawk 58]

Well, even though it's only the music scan that's failing, I can't access anything through MBC (No Content message) or Android (spinning never stops) anymore even though the web still shows content. Someone else recently posted a similar topic with no answer and this much older topic said to just delete the library (does that mean losing all watched states or are those stored separately?). I'll wait a while longer to see if I get a response before trying the delete.

[fraenhawk 58]

Oh, and sometimes the message pops up on the server and closes MBS. Other times it's just logged in the notifications but the server continues running.

[fraenhawk 58]

So I deleted the library and it was taking forever to rescan and I realized, duh, I can just restore the library from before the problem. So I did that and let it rescan the changes and everything's up and running for now. I did notice that on a lot of the music that Picard updated the tags on, I'm seeing "usicBrainz Album Id" and "usicBrainz Release Group Id" showing in the metadata manager. Some albums look like that, others are fine and show the IDs. ideas?

 

[fraenhawk 58]

Lol, I feel like I'm talking to myself, but at least this info might be useful for others. So, I looked in the tags directly and they're corrupt. So it's not MBS. I found some articles about the interaction of WMP, Windows 7, and ID3 tags with the same problem truncated problem so I just have to figure out what exactly happened and how to prevent it. Maybe it's just a matter of never using WMP again, or some people recommend setting your MP3s to read-only after Picard processes them. Regardless, I guess I have some (re-)cleanup of my whole music library.

 

The only thing I'd like to know now is what was causing MBS to competely crash? One of those bad tags being read into the database?

[porkbone 38]

I hate WMP's library service. WMP used to rewrite my album art and metadata until I removed all my mp3's from it's library. But that didn't completely stop it from making changes. I ended up having to create empty fake database files for WMP's library to finally block it from making changes. What a P.I.T.A.!

I would first remove your music from the WMP library first and see how that goes.

 

Good luck!

[fraenhawk 58]

Yeah, once I realized what was going on I took the network unc out of my music library and made sure to uncheck all the internet grabbing and renaming functions of WMP. I'll probably move to another player as well just for those times I don't listen to something through MBS. Problem is, I've still seen corruption happening and haven't narrowed down the source. At first I thought it was because I had Picard set to save the altered files to a unc so I changed it to a mapped drive letter and still saw corruption. My next tests will involve the "Enable Realtime Monitoring" option in MBS. I recall having problems years ago with Metabrowser when that was turned on and they even had notes about that function causing problems with certain file systems. I'm wondering if there's something happening that the library scan happens too quickly while Picard is still writing the file, but MBS should be read-only so I'm baffled... Most people that have reported about the Win.Explorer/WMP issue say the best thing to do is just mark all your MP3s read-only after running them through Picard, then you can still even play them in WMP without worry. I just don't know if it's strictly writing them out to the network drive or what that's causing the problem so I might have Picard retag them locally, then mark them all read-only and then finally move the library into MBS when it's all done instead.

 

I'd still love to know what caused the initial corruption where I couldn't even keep MBS running though.

 

edit: Oh, and I didn't mention I'm running WHS2011 which has it's own library sharing functions that I believe are based on WMP sharing. I turned all that off last night as well since if I'm doing remote media access I use MBS now (I used to use WHS's functions when I was using MB2).

Edited January 25, 2015 by fraenhawk

[ginjaninja 533]

 

I was getting sick of filling in all the musicbrainz artist/album/group strings manually and yet still having albums show up with no album artist so I decided to try out Picard and see if having everything embedded in the tags would help (although I would think the mediabrowser server should trump any file tags). So I started small and just had it identify 2 albums and even though Picard embedded all the tags, the metadata manager was only showing the album group after a rescan (no artist or album ID filled in). Is that normal?

 

MB3 completely relies on tags for music...in particular the output from the  ffprobe command you can see in the logs when new content is ingested.

So using picard to ensure all tags are correct is well worthwhile....but you may still need to tidy up separators etc which are compatible with your clients..

http://mediabrowser.tv/community/index.php?/topic/3100-mp3tag-actions-for-cleaning-up-collection/?p=50714

 

MB3's does not robustly and aggressively (re)acquire metadata from providors on scheduled intervals

http://mediabrowser.tv/community/index.php?/topic/17120-server-ongoing-syncing-with-metadata-providors/?p=166018

....often if it is not acquired on initial ingestion then a refresh is required to get more.

It seems normal that an MBZ album id can be known, and yet the release group and artist id which follow necessarily are not populated...not sure why this is, but since it doesnt seem to affect function, i havent worried about it..

 

 

its odd that you have text (missing Ms) populated for ids , ive never seen that....is that what actually embedded in any tags..or is this solely MB3?...does a refresh remove?

They may well cause a problem as some times these tags are passed to APIs during searches/identify/refresh functions....i would consider a fresh database/rebuild after a corruption as bad as that..

 

 

can you point me to articles of WMP corrupting tags please...I have MB3 and WMP coexisting..youve worried me..

it was necessary to turn off all internet fetching for xbox and htpc user, and changing the minimum file size for folder.jpg to prevent WMP from amending...

Edited January 25, 2015 by ginjaninja

[fraenhawk 58]

@@ginjaninja

I found out about the separators yesterday when MBZ was writing "Pop / 80s / Synth Pop" as the Genre tag (verified by MediaInfo, MP3tag, and even Windows Explorer) but ffprobe kept saying "Folk." I found some helpful articles about a bug in ffprobe related to separators causing it to prematurely end processing tags (one of them was an old post of yours! ). However the last.fm plugin works was not compatible with MB3, but the last.fm plus plugin works great, so at least I'm past that hurdle!

 

Yes, when I check the tags in anything (MediaInfo, MB3, ffprobe, MP3tag) the tag has been changed from the MBZ ID to that "usicBrainz..." so it's not MB's fault - unless it is related to the realtime scanning. Immediately reprocessing the corrupt file through MBZ seems to write it and get it to stick so it's like whatever is causing the corruption only happens when the file is created on that file system. Once it's there and just being updated it's not happening again.

 

Here are some links about WMP and non-standard ID3 tags:

http://www.smallfish-bigpond.com/2010/03/windows-media-player-12-corrupts-non-standard-id3-tags/

http://forums.musicbrainz.org/viewtopic.php?id=4173 (this one even has the missing "M" like me)

http://forums.musicbrainz.org/viewtopic.php?id=2559

http://answers.microsoft.com/en-us/windows/forum/windows_8-pictures/bug-report-explorer-and-wmp-corrupt-id3-txxx/85f1326d-9e93-4fa9-8990-ebc10501663e

 

 

[ginjaninja 533]

Thanks for links
It is surprising to hear that mb3 is reading what seem like corrupt tags for mbzid created by picard/WMP?, this (MB3 reading extended mbzid picard tags)  was talked about but i never heard it implemented.

The issue with genre tag x/y/80s/z...is that ffprobe treats 80s as well as 50s,60s,70s etc specially..... I imagine that 80s = Folk. I changed 80s tags to "The 80s".

Edited January 26, 2015 by ginjaninja

[fraenhawk 58]

Server version 3.0.5490.2

Okay, well I can provide whatever info is needed now. I copied my MP3 library from the server to my client. Removed the library from MBS. Purged the audiodb-album, audiodb-artist, fanart-music, and ffprobe-audio folders on MBS (may have been overkill, but all of these had cached info from my old music library). Configured Picard to rename and move only locally on my own PC. I then added all the ID3 tags in Picard to just a few albums/artists as a test. First, no matter what I've done on the local copy on my PC, I haven't had the tag corruption happen (yet). 

 

Test one: Copied the 5 directories over to the server share that MBS was no longer monitoring. Double-checked a few of the MP3s and wasn't seeing corruption (but didn't check every single one). Created a new music library and added the UNC to MBS and let it do its initial scan. All albums showed up, but some had the red ! triangle. Started looking through the metadata manager and noticed some of the musicbrainz tags were corrupt again. Sure enough, some of the MP3s had corrupt ID3 tags on the server only. The copies on my PC were still fine - so at least I don't have to redo Picard (again!). Wiped the MBS library, server share and cache directories again, and did a rescan.

 

Test two: set all MP3s read-only on my PC before copying them to the share. Copied them over again, double-checking a bunch of the MP3s for corrupt tags and didn't spot any. Added the library and let it rescan again. Again, seeing some with the red ! triangle on the album cover (what does that mean?), but when I start looking through the metadata manager, I don't see any of those "usicBrainz ID" corrupt tags this time. So marking read-only might be part of the solution as was mentioned in some of the articles I linked previously. Although I'd really love to know whether it's Windows Server that's corrupting the tags or MBS.

 

If I drill down to the song level, every single song now has a proper MBZ ID for artist, album artist, release group, and album, yet when I go up a level to look at the albums, some of them have the MBZ release group filled in (and TADB ids) and others have no external IDs filled in.

So problem 1: Why do some albums not get the child-level MBZ IDs copied up to them

 

Problem 2: I have my music album and music artist metadata downloaders both configured to MBZ before TADB, yet if TADB doesn't have an album description but the MBZ release group does, the overview remains blank. Sounds like the MBZ fetcher doesn't support overview. That's a shame since MBZ has way more identified albums than TADB.

 

Problem 3: Even when the child MBZ ID gets copied from the song to the album, only the release group gets copied (most of the time - see #4). The artist and album ID remain blank even though every song had identical MBZ IDs. The Album ID from the song should populate the album (just like the release group ID) and the song's album artist ID should copy to the album's artist ID (this prevents reconciling different artists per song since the album artist should be the same).

 

Problem 4: Sometimes, everything does get copied up from the song to the album. My guess is it's only when I only have one song in the album. I think there is a problem with (or lack of) reconciliation for when there are multiple songs in an album.

Here's an example of where even the single song approach isn't working though. I have a "Various Artists" album. The song is properly labeled with the ID for "Various Artists" in the album artist, and the true artist as the artist ID. The artist for the song got copied into the artist on the album instead of the album artist. But then, the "Various Artist" at the artist level of the metadata manager has the proper VA ID. What this leads to is some major confusion in the web interface when browsing music. Both the "real" artist and the album artist show up in the "Artists" tab of music on the web. Well VA never had a song in its name so shouldn't show up at all there, only in the Album Artist tab. When you click on either artist, you get the same album and song, but the song shows the correct true artist on the web (oh, and there's another one of those ! triangles). If I go to Album Artists, the real artist does not show up (good).

I think what might be happening is that the song's album artist is copied all the way up to the artist ID of the artist, but not being used as the artist ID of the album also. My reasoning is that I have a multi-album artist (slash delimited) which I'm pretty sure I've read another thread about MBS not yet supporting, but it did copy the pair of album artist IDs up to the combined artist that it's showing, but the artist on the album is blank.

 

Problem 5: Caching from previous library (?). Out of the 5 artists I added for these tests, one of them has "usicBrainz Album Artist Id" in the external MusicBrainz ID. This artist had two albums, each with only 1 MP3 in them. I checked the metadata on both MP3s and both are showing proper MBZ IDs for every ID3 tag at the song level. Both albums show no IDs due to problem #1 above. So the only place I can think that it's getting the corrupt tag is from a previous version of the library, right?

 

I know video is definitely a higher priority than audio, but I love the fact that MBS supports MBZ IDs and I feel has a nice browsing playback interface - for both the web, and MBC - but if I can't just have it read all the folders/files without micromanaging everything, I might as well go back to letting WHS serve up my music through it's web sharing interface.

 

I can provide whatever logs, screenshots, ffprobe reads that you'd like if someone has the time to work on the music side of MBS. I really just feel that now that I'm past the corruption by setting read-only, this is just a matter of which fields are read for what purposes and just a few tweaks to the logic & fetchers would go a long way.

[fraenhawk 58]

Since I haven't gotten any feedback or requests for logs, I went ahead and started working with those few albums I've added back into the library. I have plenty more that can be moved in for fresh logs/experiments. One of my concerns was if I copied the album, release, and album artist from the tracks to the albums that a refresh might undo all that work. Luckily that isn't happening.

 

The main overview is definitely tied to theaudiodb like I reported even though MBZ is my primary metadata collector in MBS options.

 

And, I think I finally figured out what the annoying red ! triangles are on the album covers. It means that there is no TADB ID configured for that album. I'm not sure why so much emphasis was placed on TADB in MBS when MBZ seems to have all the information you could need, and usually more.

[fraenhawk 58]

I think I found a bug with the refresh with a third scenario.

Scenario 1: MBZ and TADB set, overview is pulled in from TADB. Refresh does nothing wrong.

Scenario 2: MBZ IDs set, album not found in TADB. I manually copy the overview from MBZ. Refresh does not remove any manually added metadata.

Scenario 3: MBZ ID and TADB IDs set, but TADB doesn't have an overview. I manually copy the overview from MBZ. Refresh wipes out the overview. Even using Advanced Refresh with it set to Add Missing Data Only wipes out the overview. In my mind, that's bug. Unless I'm missing something, that setting should only be touching null metadata elements, correct?

[fraenhawk 58]

Ugh, and now I've run into the infamous "multiple artists" dilemma that I've seen other threads mentioning. Is the build that uses Artists (vs Artist) in beta yet?

If anyone has any advice for handling multi-artist albums that would be appreciated.

 

My specific example is the collaborative album, Peter & the Wolf from "Weird Al" Yankovic & Wendy Carlos.

Picard places it in a new artist folder of "_Weird Al_ Yankovic & Wendy Carlos"

It sets the Artist ID to "7746d775-9550-4360-b8d5-c37bd448ce01/7e3b3173-2c1f-42c0-8544-0e6d19bb181f"

That's my first question, does semi-colon work better than slash here or are neither supported yet?

So MBS appears to read the first tag and in the web interface shows Weird Al twice in the Album Artists and Artists section. Clicking on the first one gives all the albums for the singular Weird Al. Clicking on the second gives no albums at all.

Album view shows the album and clicking on it allows playing the songs, and the artist shows as "Weird Al" Yankovic & Wendy Carlos. If I click the artist link from the album, I go to the Weird Al description. Oh, and if I click edit on the one with no albums/songs, I go to the metadata manager with an image by name/artist path, rather than either artist folder. I think the problem is the & in the name (related to this?).

 

The thing is, no matter how much I manipulate the metadata it seems MBS is going by the path name or maybe the song artist metadata?? I tried setting the album artist to just Weird Al's MBZ ID, I tried Wendy Carlos' ID, I set the display name to just one or the other and no matter what, the album still shows the two artists with the &

[fraenhawk 58]

Aha, I think I've figured out that the image by name/artist entries are created for when MBS doesn't find an artist folder. So if you have an album for artist A, but artist B sings on it but you have no artist B albums, the metadata edit link takes you to the imagebyname entry. Any album artist I have that matches an artist folder has its images saved in the music library folder, any unmatched gets its images saved in imagesbyname.

 

That that leads me to believe then is that MBS is having problems with any artist with special characters. For instance, I have AC/DC albums. You can't have a slash in the folder name so Picard creates an "AC_DC" folder. If I open my music\ac_DC folder I see 4 image files that MBS put in there. If I look in ImagesByName/artists, I have an "AC DC" folder (space instead of underscore). In the metadata manager, I deleted the banner. It deleted from the AC_DC folder. If I go into an AC/DC album, click the artist link, then edit, I get taken to the imagesbyname version, not the \\music\ac_dc version of the artist.

 

Other artists that don't have special characters (Taylor Swift) only have their images in the \music\Taylor Swift folder and don't have an imagesbyname/artists folder at all. If I go into one of her albums and click her name, then edit, I get taken to \music\taylor swift, not imagesbyname in the metadata manager.

 

Picard creates "Weird Al" Yankovic as "_Weird Al_ Yankovic" but an imagesbyname/artists folder of "weird al  yankovic" gets created (yes there are two spaces between al and yankovic - is that because it's replacing the _ with space and you can't start a folder with space?)

Edited February 5, 2015 by fraenhawk

[fraenhawk 58]

And, just to add to my problems list. I'm also seeing instances (not sure if changes were made with the newest beta server or just hadn't run into this) where the album ID is populated, but with some other album ID. It didn't come from any of the songs. This has happened quite a few times on the next small batch of albums I've added back into my music library.

 

Specific example: Adele's album 19. Every song was labeled by Picard with album ID aea26ecc-887a-47a6-8b35-069ae727a6ed

yet when it was added back into the library (copied to server, forced a re-scan), the album ID showing at the album level in metadata manager was 470e4c8b-f156-4dc5-ab4c-adb726f95b65. That's not even another version of the album from the same release group!

 

Oh, and the release group showing was 549ae43d-59cf-4bc8-84ec-339b1cf9c1c8 instead of 9796da06-2d59-3176-8598-2105f31ee54a so at least I know where it got the album from. But where did it get the release group... part of the caching problem I've noticed or something unrelated? I verified that MBS read the same ID ending in 54a on every single track so I'm not sure where that 1c8 came from.

 

It's bad enough that for nearly every album I'm having to copy and paste data from the song level up to the album level, but to have to question the data that is showing is even worse. Now, if I don't remember if I already fixed an album, I have to go double-check all the metadata. I know video is probably a higher priority than audio for the team, but the lack of response to this topic is surprising. Usually some official response is pretty quick, even if it's something like "we're aware of the problems but aren't actively working on them."

[fraenhawk 58]

Definitely can confirm caching. MusicBrainz labels the earlier Black Eyed Peas albums without the "The" and later ones with "The" in the artist name. I didn't like having them listed twice in the Album Artists view (once each way) with 3 albums attributed to one version and 2 to the other. So I decided to copy them all into the folder with the "The" and rescan the library. The metadata manager showed them all combined into one artist, but the web interface still showed two versions. Problem now was that whichever you clicked on, you only got the 3 albums that were originally showing up with the "The"

If you looked by album, the other two were there and if you clicked on them, then clicked on the artist, you wound up back on the version that didn't have the album you just came from. I even went through the trouble of amending the tags in musicbrainz so that all 5 albums were consistent - always showing the "The" for the artist and album artist. Still no luck after rescan. Tried removing the whole artist directory, rescan so it knew they were gone, then put them back in and rescan. Still no luck. Finally, to verify this is a caching issue I installed SQLite Studio and sure enough I found lots of caching problems. There were references to the old path that didn't exist anymore as well as a copy of the album and tracks that had the artist without the "The". Shutdown the server, removed those rows, started up and rescanned and everything is finally working correctly!

 

The problem is that there is data stored on the Album that is not shown in the metadata manager. Metadata manager does not show the album artist, only the musicbrainz and tadb IDs (which were correct), but behind the scenes in the row itself was an incorrect album artist. And the rescan isn't updating those values based on the ID3 tag changes.