Motivation:
Sometimes you need to index a bunch of documents really, really fast. Even with Solr 4.0 and soft commits, if you send one document at a time you will be limited by the network. The solution is two-fold: batching and multi-threading. http://lucidworks.com/blog/high-throughput-indexing-in-solr/
solrbulk expects as input a file with line-delimited JSON. Each line represents a single document. solrbulk takes care of reformatting the documents into the bulk JSON format, that SOLR understands.
solrbulk will send documents in batches and in parallel. The number of
documents per batch can be set via -size
, the number of workers with -w
.
This tool has been developed for project finc at Leipzig University Library.
Installation via Go tools.
$ go install github.com/miku/solrbulk/cmd/solrbulk@latest
There are also DEB and RPM packages available at https://github.com/miku/solrbulk/releases/.
Flags.
$ solrbulk
Usage of solrbulk:
-commit int
commit after this many docs (default 1000000)
-cpuprofile string
write cpu profile to file
-memprofile string
write heap profile to file
-no-final-commit
omit final commit
-optimize
optimize index
-purge
remove documents from index before indexing (use purge-query to selectively clean)
-purge-pause duration
insert a short pause after purge (default 2s)
-purge-query string
query to use, when purging (default "*:*")
-server string
url to SOLR server, including host, port and path to collection,
e.g. http://localhost:8983/solr/biblio
-size int
bulk batch size (default 1000)
-update-request-handler-name string
where solr.UpdateRequestHandler is mounted on the server,
https://is.gd/s0eirv (default "/update")
-v prints current program version
-verbose
output basic progress
-w int
number of workers to use (default 4)
-z unzip gz'd file on the fly
Given a newline delimited JSON file:
$ cat file.ldj
{"id": "1", "state": "Alaska"}
{"id": "2", "state": "California"}
{"id": "3", "state": "Oregon"}
...
$ solrbulk -verbose -server https://192.168.1.222:8085/collection1 file.ldj
The server parameter contains host, port and path up to, but excluding the
default update
route
for search (since 0.3.4, this can be adjusted via
-update-request-handler-name
flag).
For example, if you usually update via https://192.168.1.222:8085/solr/biblio/update
the server parameter would be:
$ solrbulk -server https://192.168.1.222:8085/solr/biblio file.ldj
- Having as many workers as core is generally a good idea. However the returns seem to diminish fast with more cores.
- Disable
autoCommit
,autoSoftCommit
and the transaction log insolrconfig.xml
. - Use some high number for
-commit
. solrbulk will issue a final commit request at the end of the processing anyway. - For some use cases, the bulk indexing approach is about twice as fast as a standard request to
/solr/update
. - On machines with more cores, try to increase maxIndexingThreads.
Try esbulk.