Lucene config for docker?

Leo · May 10, 2019, 1:27am

Hi Sebastian, how do we persist Lucene data when Squidex is running on a docker container? The CMS is deployed in GKE and everytime the pod restarts we get nothing from search queries

Sebastian · May 10, 2019, 5:55am

Hi, The lucene index is stored in the asset store and then extracted to the temp folder. This was the only solution I found that would work with kubernetes.

github.com

Squidex/squidex/blob/master/src/Squidex.Domain.Apps.Entities/Contents/Text/TextIndexerGrain.cs#L63




public override async Task OnDeactivateAsync()
{
    await DeactivateAsync(true);
}


protected override async Task OnActivateAsync(Guid key)
{
    directory = new DirectoryInfo(Path.Combine(Path.GetTempPath(), $"Index_{key}"));


    await assetStore.DownloadAsync(directory);


    var config = new IndexWriterConfig(Version, Analyzer)
    {
        IndexDeletionPolicy = snapshotter,
        MergePolicy = new TieredMergePolicy(),
        MergeScheduler = MergeScheduler
    };


    indexWriter = new IndexWriter(FSDirectory.Open(directory), config);

Junim · July 13, 2020, 7:49pm

Hi, @Sebastian. Is this still the current solution?

Sebastian · July 13, 2020, 8:08pm

Yes, it is stored in the assets. In the meantime I tried other solutions, but this is the fastest.

Why do you ask?

Junim · July 13, 2020, 8:31pm

Tks for reply. I’ve made this question because I’m facing a similar problem (using AKS). There is no errors in logs so I’m investigating a possible problem with docker volume.

Sebastian · July 14, 2020, 4:48am

Just use S3 as asset storage…

Junim · July 14, 2020, 4:13pm

I’m analyzing the code where the temp folder is deleted. When removing the IF block, Lucene data is no longer removed and the search continues to work even after an deploy. I’m trying to understand this part better and why this deletion is always done in the first search after running the application.

github.com

Squidex/squidex/blob/master/backend/src/Squidex.Domain.Apps.Entities.MongoDb/FullText/MongoIndexStorage.cs#L39


}


public async Task<LuceneDirectory> CreateDirectoryAsync(DomainId ownerId)
{
    var fileId = $"index_{ownerId}";


    var directoryInfo = new DirectoryInfo(Path.Combine(Path.GetTempPath(), fileId));


    if (directoryInfo.Exists)
    {
        directoryInfo.Delete(true);
    }


    directoryInfo.Create();


    try
    {
        using (var stream = await bucket.OpenDownloadStreamAsync(fileId))
        {
            using (var zipArchive = new ZipArchive(stream, ZipArchiveMode.Read, true))
            {

Sebastian · July 14, 2020, 4:20pm

The lucene indexes are implemented as an actor. An actor is a stateful instance that is distributed across a Squidex cluster by Orleans (https://dotnet.github.io/orleans/Documentation/index.html).

Each app has their own actor for full text search.

It is like a set of microservices living inside Squidex. When you have multiple nodes and one node goes down, is restarted or a new version is deployed the actors on this node are destroyed and recreated when needed either on the same machine or one of the other machines.

Therefore an index can move from machine A to machine B and back to machine A and we cannot store the indexes in the file system of the machine. Therefore they are stored in the asset storage, which could be a file system but is usually S3 bucket or any other cloud storage.

After this transition A->B->A, the index in the temp folder on A might be stale, but I don’t know it. Therefore I delete it and download the newer version from the asset store.

Junim · July 14, 2020, 4:32pm

Tks for quick and complete response.

Sebastian · July 14, 2020, 4:36pm

The point is: The full text search uses the same storage as the assets. You want to have it fixed anyway in your deployment. Otherwise you will loose your assets as well.