1.4 Configuration
After installing, a directory structure will be created in the install directory. In it, the following files and directories will be created:
| path | description |
|---|---|
| conf/properties | java-style properties files for configuring the search engine |
| conf/suggestions.txt | A list of suggestions for the search interface |
| conf/db.properties | JDBC configuration |
| env/mariadb.env | Environment variables for the mariadb container |
| env/service.env | Environment variables for Marginalia Search services |
| logs/ | Log files |
| model/ | Language processing models |
| index-1/backup | Index backups for index node 1 |
| index-1/index | Index data for index node 1 |
| index-1/storage | Raw and processed crawl data for index node 1 |
| index-1/work | Temporary work directory for index node 1 |
| index-1/uploads | Upload directory for index node 1 |
| index-2/backup | Index backups for index node 2 |
| index-2/index | Index data for index node 2 |
| … | … |
For a production-like deployment, you will probably want to move the db and index directories to a separate storage device. The index-n/storage directory wants as much space as can be afforded, and index-n/index wants an SSD.
Recommended Tweaks for constrained hardware
If you intend to run the system on a machine with limited RAM, especially on a relatively high-powered CPU with many cores, some configuration changes are recommended.
In conf/properties/system.properties:
-
Add
system.conserveMemory = true. This will make the system use additional memory conservation strategies where available, at the expense of disk I/O and speed. -
Optionally, you might want to tweak
converter.poolSize. The default is automatically reduced bysystem.conserveMemory, and affects the number of threads used by the converter process in turn increasing RAM requirements. It should be in the range between 1 and the number of logical cores in your CPU. If set lower than necessary processing will take longer time. -
You might also want to lower
converter.sideloadThresholdto something smaller than the default 10,000 – maybe down to 1000. This will make the converter perform a simpler processing operation on domains with more than this number of documents. It’s generally a fair bit slower, but drastically reduces the amount of RAM required.
Optional:
By default, Java sets a maximum heap size of 25% of your system RAM. In general this is fine, but in extremely resource constrained systems, this is a bit low.
If ${MEMORY} is half your system ram in Gb, edit env/service.env and add -Xmx${MEMORY}Gb to CONVERTER_PROCESS_OPTS.
That is, if you have 32 GB of RAM, it could look like this:
CONVERTER_PROCESS_OPTS="-Dservice-name=converter -Dservice-host=0.0.0.0 -Xmx16G"
This is not generally necessary.
system.properties overview
Of principal importance when configuring the system is conf/properties/system.properties, which is the main configuration file for the search engine. It contains the following properties:
Global
| flag | values | description |
|---|---|---|
| flyway.disable | boolean | Disables automatic Flyway database migrations |
| log4j2.configurationFile | string | log4j2-test.xml, log4j2-prod.xml |
| system.conserveMemory | boolean | Apply extra effort to conserve RAM. Useful on low-powered hardware |
Crawler Properties
| flag | values | description |
|---|---|---|
| crawler.userAgentString | string | Sets the user agent string used by the crawler |
| crawler.userAgentIdentifier | string | Sets the user agent identifier used by the crawler, e.g. what it looks for in robots.txt |
| crawler.poolSize | integer | Sets the number of threads used by the crawler, more is faster, but uses more RAM. This should generably higher than the number of cores, since crawler threads idle most of the time. |
| crawler.initialUrlsPerDomain | integer | Sets the initial number of URLs to crawl per domain (when crawling from spec) |
| crawler.maxUrlsPerDomain | integer | Sets the maximum number of URLs to crawl per domain (when recrawling) |
| crawler.minUrlsPerDomain | integer | Sets the minimum number of URLs to crawl per domain (when recrawling) |
| crawler.crawlSetGrowthFactor | double | If 100 documents were fetched last crawl, increase the goal to 100 x (this value) this time |
| ip-blocklist.disabled | boolean | Disables the IP blocklist |
Converter Properties
| flag | values | description |
|---|---|---|
| converter.sideloadThreshold | integer | Threshold value, in number of documents per domain, where a simpler processing method is used which uses less RAM. 10,000 is a good value for ~32GB RAM |
| converter.poolSize | integer | Number of threads used by the converter process, should probably be in the range 2 - 32 |
Marginalia Application Specific
| flag | values | description |
|---|---|---|
| control.hideMarginaliaApp | boolean | Hides the Marginalia application from the control GUI (used in barebones mode) |
Experimental Flags (DANGER!)
These flags are provisional and may have unintended side effects. These are subject to change, migration paths are not guaranteed.
| flag | values | description |
|---|---|---|
| system.noSunMiscUnsafe | boolean | Disable the use of sun.misc.Unsafe, use MemorySegment instead. This is slower but has better memory safety |
| system.languageDetectionModelVersion | integer | Choose language detection model. -1 to disable, 0 for default; 1 or 2 to select model between the old crude model and fasttext (which despite its name is slower but more accurate). |
| system.noFlattenUnicode | boolean | if true, the search engine will attempt to support utf-8 keywords. The index needs to be re-constructed after this has been switched. Probably also needs new language models. |