reindexer

TRLN Discovery Indexer

Standalone tool for (re)indexing TRLN Discovery documents from the shared database into a Solr collection.

Useful to prepare a new Solr instance or SolrCloud cluster with documents that expect a different structure than the current production index.

Details

It runs a query against the "Spofford" database to pull out the Argot of all current documents, outputs those into files of a (configurable) number of records, then runs argot ingest on the resulting files. (see the argot-ruby gem for more information on this process). This has the effect of reindexing all the documents that match the original query.

Because this might involve a lot of documents and take some time, the driver is written in Go and uses concurrent workers for the argot and Solr-related parts of the process.

This process is designed primarily for indexing into a non-production collection, to test new index configurations, or prepare for Solr upgrades. It may involve the use of different version of Argot or Solr than the ones used in production. If you want to reindex things in production, you are strongly encouraged to use or adapt tools available in trln-ingest (spoffford) for that.

Building

If you are running this on either Amazon Linux or some flavour of Red Hat Enterprise Linux (including CentOS and Fedora), you should be able to install all the prequisites and build the driver program by simply running

$ ./init.sh [optional argot branch]

In this directory. If you're not running one of those flavours of Linux, use the cues in that script as a starting point (mostly that will involve the name of the OS' package manager and the names of the packages).

The build process will also pull down argot-ruby and build it; if you want to use any version of argot other than the one in the master branch of that repository, pass in the branch name as the argument to the init script.

As a final step, the script builds the go program and installs it to ~/bin/reindexer (as well as to ~/go/bin/reindexer; the latter is less likely to be on your path)

Running

Once built, the reindexer program can be copied anywhere on the system and run from there. It takes one optional argument, which is the name of a configuration file (format described below). If omitted, it will load the configuration from config.json in the current working directory.

Configuration

The configuration file's format is JSON, and has the following structure (see config/config.go and the definition of the Config struct for more guidance):

{
    "host" : "localhost",
    "port" : 5432,
    "database" : "shrindex",
    "user": "shrindex",
    "password" : "no default",
    "query" : "SQL query used to fetch documents",
    "chunkSize" : 20000,
    "solrUrl" : "http://localhost:8983/solr/trlnbib",
    "authorities": true,
    "redisUrl": "redis://localhost:6379/0",
    "workers" : 3,
    "startId": ""
}

Where the value in the sample above looks like a sensible value, it's the default; you MUST provide at least password. workers defaults to the number simultaneous threads the current machine can run (never lower than 1). The default value for query selects all non-deleted documents from the database, and documents are retrieved in "asciibetical" sort order.

startId is provided in case the reindex process is interrupted, and you need to start over; when set to a non-empty string, all documents where the id field sorts after this value will be indexed.

You can query the Solr index for the 'last' ingested document ID via the following parameters to the select handler:

$ curl -s 'http://localhost:8983/solr/gettingstarted/select?q=*:*&fl=id&sort=id+desc&rows=1' | jq .response.docs[0].id`

(jq is installed by default by the ansible script used to install this project).

Authority processing against a redis database (the authorities attribute) is true by default. When authorities is true, the redisUrl will be checked to see whether it's accessible, and if this check fails, the process will abort.

The master process loops over the documents matching the query, and outputs them into files with no more than chunkSize records in them. At that point, it passes that file to an available worker, which then runs argot ingest -s [solrUlr] -a --redis-url [redisUrl] on the file (assuming authorities is true; if false omit the -a, --redis-url and redis URL parameters), which flattens and suffixes the Argot records in the file, and then submits the results to Solr for reindex.

The entire process is logged to STDERR, so you may want to run the driver program thusly:

$ ~/go/bin/reindexer 2> ingest-log-$(date +%Y-%m-%d).log&

To run it in the background, and then poke into the log now and again to check on progress. Prepend this command with nohup to allow you to log off or get disconnected without terminating the process.