Search

The search service is responsible for metadata and content extraction, stores that data as index and makes it searchable. The following clarifies the extraction terms metadata and content:

• Metadata: all data that describes the file like Name, Size, MimeType, Tags and Mtime.
• Content: all data that relates to content of the file like words, geo data, exif data etc.

• General Considerations
• Search engines
• Query language
• Extraction Engines
• Content Extraction
  • Basic Extractor
  • Tika Extractor
• Search Functionality
  • Indexing
  • Search
  • State Changes which Trigger Indexing
    • Resource Trashed
    • Resource Deleted
    • Resource Restored
    • Resource Moved
    • Folder Created
    • File Created
    • File Version Restored
    • Resource Tag Added
    • Resource Tag Removed
    • File Uploaded - Synchronous
    • File Uploaded - Asynchronous
• Manually Trigger Re-Indexing a Space
• Notes
• Example Yaml Config

• To use the search service, an event system needs to be configured for all services like NATS, which is shipped and preconfigured.
• The search service consumes events and does not block other tasks.
• When looking for content extraction, Apache Tika - a content analysis toolkit can be used but needs to be installed separately.

Extractions are stored as index via the search service. Consider that indexing requires adequate storage capacity - and the space requirement will grow. To avoid filling up the filesystem with the index and rendering Infinite Scale unusable, the index should reside on its own filesystem.

You can change the path to where search maintains its data in case the filesystem gets close to full and you need to relocate the data. Stop the service, move the data, reconfigure the path in the environment variable and restart the service.

When using content extraction, more resources and time are needed, because the content of the file needs to be analyzed. This is especially true for big and multiple concurrent files.

The search service runs out of the box with the shipped default basic configuration. No further configuration is needed, except when using content extraction.

Note that as of now, the search service can not be scaled. Consider using a dedicated hardware for this service in case more resources are needed.

By default, the search service is shipped with bleve as its primary search engine. The available engines can be extended by implementing the Engine interface and making that engine available.

By default, KQL is used as query language, for an overview of how the syntax works, please read the microsoft documentation.

Not all parts are supported, the following list gives an overview of parts that are not implemented yet:

• Synonym operators
• Inclusion and exclusion operators
• Dynamic ranking operator
• ONEAR operator
• NEAR operator
• Date intervals

In the following ADR you can read why we chose KQL.

The search service provides the following extraction engines and their results are used as index for searching:

• The embedded basic configuration provides metadata extraction which is always on.
• The tika configuration, which additionally provides content extraction, if installed and configured.

The search service is able to manage and retrieve many types of information. For this purpose the following content extractors are included:

This extractor is the most simple one and just uses the resource information provided by Infinite Scale. It does not do any further analysis. The following fields are included in the index: Name, Size, MimeType, Tags, Mtime.

This extractor is more advanced compared to the Basic extractor. The main difference is that this extractor is able to search file contents. However, Apache Tika is required for this task. Read the Getting Started with Apache Tika guide on how to install and run Tika or use a ready to run Tika container. See the Tika container usage document for a quickstart. Note that at the time of writing, containers are only available for the amd64 platform.

As soon as Tika is installed and accessible, the search service must be configured for the use with Tika. The following settings must be set:

• SEARCH_EXTRACTOR_TYPE=tika
• SEARCH_EXTRACTOR_TIKA_TIKA_URL=http://YOUR-TIKA.URL

When the search service can reach Tika, it begins to read out the content on demand. Note that files must be downloaded during the process, which can lead to delays with larger documents.

Content extraction and handling the extracted content can be very resource intensive. Content extraction is therefore limited to files with a certain file size. The default limit is 20MB and can be configured using the SEARCH_CONTENT_EXTRACTION_SIZE_LIMIT variable.

When extracting content, you can specify whether stop words like I, you, the are ignored or not. Noramlly, these stop words are removed automatically. To keep them, the environment variable SEARCH_EXTRACTOR_TIKA_CLEAN_STOP_WORDS must be set to false.

When using the Tika container and docker-compose, consider the following:

• See the ocis_wopi example.
• Containers for the linked service are reachable at a hostname identical to the alias or the service name if no alias was specified.

If using the tika extractor, make sure to also set FRONTEND_FULL_TEXT_SEARCH_ENABLED in the frontend service to true. This will tell the webclient that full-text search has been enabled.

The search service consists of two main parts which are file indexing and file search.

Every time a resource changes its state, a corresponding event is triggered. Based on the event, the search service processes the file and adds the result to its index. There are a few more steps between accepting the file and updating the index.

A query via the search service will return results based on the index created.

The following state changes in the life cycle of a file can trigger the creation of an index or an update:

The service checks its index to see if the file has been processed. If an index entry exists, the index will be marked as deleted. In consequence, the file won’t appear in search requests anymore. The index entry stays intact and could be restored via Resource Restored.

The service checks its index to see if the file has been processed. If an index entry exists, the index will be finally deleted. In consequence, the file won’t appear in search requests anymore.

This step is the counterpart of Resource Trashed. When a file is deleted, is isn’t removed from the index, instead the service just marks it as deleted. This mark is removed when the file has been restored, and it shows up in search results again.

This comes into play whenever a file or folder is renamed or moved. The search index then updates the resource location path or starts indexing if no index has been created so far for all items affected. See Notes for an example.

The creation of a folder always triggers indexing. The search service extracts all necessary information and stores it in the search index

This case is similar to Folder created with the difference that a file can contain far more valuable information. This gets interesting but time-consuming when data content needs to be analyzed and indexed. Content extraction is part of the search service if configured.

Since Infinite Scale is capable of storing multiple versions of the same file, the search service also needs to take care of those versions. When a file version is restored, the service starts to extract all needed information, creates the index and makes the file discoverable.

Whenever a resource gets a new tag, the service takes care of it and makes that resource discoverable by the tag.

This is the counterpart of Resource tag added. It takes care that a tag gets unassigned from the referenced resource.

This case only triggers indexing if async post processing is disabled. If so, the service starts to extract all needed file information, stores it in the index and makes it discoverable.

This is exactly the same as File uploaded - synchronous with the only difference that it is used for asynchronous uploads.

The service includes a command-line interface to trigger re-indexing a space:

ocis search index --space $SPACE_ID --user $USER_ID

Note that not names but IDs are necessary and that the specified user ID needs access to the space to be indexed.

The indexing process tries to be self-healing in some situations.

In the following example, let’s assume a file tree foo/bar/baz exists. If the folder bar gets renamed to new-bar, the path to baz is no longer foo/bar/baz but foo/new-bar/baz. The search service checks the change and either just updates the path in the index or creates a new index for all items affected if none was present.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54

# Autogenerated
# Filename: search-config-example.yaml

tracing:
  enabled: false
  type: ""
  endpoint: ""
  collector: ""
log:
  level: ""
  pretty: false
  color: false
  file: ""
debug:
  addr: 127.0.0.1:9224
  token: ""
  pprof: false
  zpages: false
grpc:
  addr: 127.0.0.1:9220
  tls: null
token_manager:
  jwt_secret: ""
reva:
  address: com.owncloud.api.gateway
  tls:
    mode: ""
    cacert: ""
grpc_client_tls: null
events:
  endpoint: 127.0.0.1:9233
  cluster: ocis-cluster
  async_uploads: true
  num_consumers: 0
  debounce_duration: 1000
  tls_insecure: false
  tls_root_ca_certificate: ""
  enable_tls: false
  username: ""
  password: ""
engine:
  type: bleve
  bleve:
    data_path: /var/lib/ocis/search
extractor:
  type: basic
  cs3_allow_insecure: false
  tika:
    tika_url: http://127.0.0.1:9998
    clean_stop_words: true
content_extraction_size_limit: 20971520
service_account:
  service_account_id: ""
  service_account_secret: ""