Well Valerio, my expertise is in ontology and knowledge representation, so I hope not to give you wrong information.
I do not have a back up, because I have not finished the first load of data, which it is a 1,5 M items, around 60 M of triples.
At the beginning I realized that increasing threads increased the edition, with threads we increased edition to 4 items/second, then we improved hardware, and with the sysadmin made the following adjustment to our database:
For database:
We heave test with mariadb and mysql Both we have set in tmpfs (temporary file storage) and done following settings in mysqld.cnf
tmpdir = /var/lib/mysql/mysqltmp query_cache_limit = 0 query_cache_size = 0 innodb_buffer_pool_size = 8G innodb_flush_log_at_trx_commit=2
With this improvement we reached our max edition 10 items/second.
However, we do not know how to make further improvement, given that we must load 30 M items, which seems to be very long task.
Hope this information could shade some lights to our use case, and perhaps helps us to improve our work.
Luis
Valerio Bozzolan boz+wiki@reyboz.it hat am 29. Januar 2020 um 08:44 geschrieben:
Thank you for the clarification,
First of all let me clarify that on you private Wikibase instance - on your own hardware - you can surely do whatever you want and flood your APIs without asking any permission. So, if you reached a pratical edit/second limitation, probably you may want to find some hardware bottlenecks with the help of a sysadmin.
As a note "in case of fire" you can just restore your database backup instead of re-running your bot another time. (You have a backup, isn't it? :)
Warm wishes
On January 29, 2020 8:12:40 AM GMT+01:00, wp1080397-lsrs wp1080397-lsrs luis.ramos@pst.ag wrote:
If I understand your request, I can not provide you such discussion, because I did not participate in any discussion for bot approval, our administrator configured the bots in our private instance.
Hope you can provide me some additional support, and
if you require further information, please let me now, and I would answer ASAP.
Best regards
Luis Ramos
Valerio Bozzolan boz+wiki@reyboz.it hat am 28. Januar 2020 um 17:43
geschrieben:
In order to further help you, can I ask you your Wikidata bot
approval
discussion?
https://www.wikidata.org/wiki/Wikidata:Requests_for_permissions/Bot
On Tue, 2020-01-28 at 10:19 +0100, wp1080397-lsrs wp1080397-lsrs
wrote:
Dear Valerio,
Thanks for the quick answer, if I understood your answer, we should be using an inappropriate approach at doing parallel programming in the edition process. In this case, we are aiming to have the data available asap, as
soon
as we have it we should use another approach.
The question I made is about the necessity of loading large data sets, because in the case of private instances, we need to load 20.000.000 of items for private use, and with a rate of 10 items
per
second, using the approach we are following we will require 25
days,
with a script writing 24 hour a day, and speaking in big data
terms,
20 M is an small data set.
So, I leave an open question:
my questions is if there is some experience when has been possible
to
have a higher speed in edition rate?.
Best regards
Valerio Bozzolan boz+wiki@reyboz.it hat am 28. Januar 2020 um 09:28 geschrieben:
Please note that - AFAIK - parallel requests are not well
accepted.
https://www.mediawiki.org/wiki/API:Etiquette
(You may have a bigger problem now :^)
On Tue, 2020-01-28 at 08:13 +0100, wp1080397-lsrs wp1080397-lsrs wrote:
Dear friends, We have been working for some months in a wikidata project, and we have found an issue with edition performance, I began to work with wikidata java api, and when I tried to increase the edition
speed
the java system held editions, and inserted delays, which reduced edition output as well. I chose the option to edit with pywikibot, but my experience
was
that this reduced more the edition. At the end we use the procedure indicated here: https://www.mediawiki.org/wiki/API:Edit#Example With multithreading, and we reach a maximum of 10,6 edition per second. my questions is if there is some experience when has been possible to have a higher speed?. Currently we need to write 1.500.000 items, and we would
require
5 working days for such a task. Best regards Luis Ramos Senior Java Developer (Semantic Web Developer) PST.AG Jena, Germany.
Mediawiki-api mailing list Mediawiki-api@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-api
Mediawiki-api mailing list Mediawiki-api@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-api
Luis Ramos
Senior Java Developer
(Semantic Web Developer)
PST.AG
Jena, Germany.
Mediawiki-api mailing list Mediawiki-api@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-api
Mediawiki-api mailing list Mediawiki-api@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-api
Luis Ramos
Senior Java Developer
(Semantic Web Developer)
PST.AG
Jena, Germany.
-- E-mail sent from the "K-9 mail" app from F-Droid, installed in my LineageOS device without proprietary Google apps. I'm delivering through my Postfix mailserver installed in a Debian GNU/Linux.
Have fun with software freedom!
[[User:Valerio Bozzolan]]
Mediawiki-api mailing list Mediawiki-api@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-api
Luis Ramos
Senior Java Developer
(Semantic Web Developer)
PST.AG
Jena, Germany.