@mcscx2 @lakwnikos don't think there is a ddos. A classical slow query is the following example (examine 3mill rows). "# Query_time: 16.042286 Lock_time: 0.000140 Rows_sent: 104 Rows_examined: 3776090 # Rows_affected: 0 SET timestamp=1527533585; SELECT id FROM notice WHERE ( notice.created > "2018-01-14 23:36:01" ) AND ( notice.id IN (SELECT notice_id FROM reply WHERE profile_id=45991) OR notice.profile_id IN (SELECT subscribed FROM subscription WHERE subscriber=45991) OR notice.id IN (SELECT notice_id FROM group_inbox WHERE group_id IN (SELECT group_id FROM group_member WHERE profile_id=45991))OR notice.id IN (SELECT notice_id FROM attention WHERE profile_id=45991) ) ORDER BY notice.id DESC LIMIT 0, 200;" Also query's from the search field do a "LIKE '%yoursearch%" - which takes a long time - like does not use indexes. However, the select above (example 1) is the one reporting often as the slow queries.
@knuthollund maybe other !gnusocial admins could check one their instances whether they have those slow queries, too.
Because I wonder why this seems to affect only quitter.no (apart from also quitter.se which is currently down), _even though_ quitter.no has already such powerful hardware like SSD, 12 GB RAM and the whole database kept in memory. I would expect there are at least some less powerful instances around, but I'm still still havent found other instances affected by this #every-few-minutes-unresponsive-for30-seconds-issue.
@knuthollund interesting. Is that a query for an id of a notice which has been created after "2018-01-14 23:36:01" (plus some other criteria)? I wonder if that is already strange?
How often does such a slow query occur? In practise the non-responding conditions come every 2-5 minutes and always seem to last for ~29 or 30 seconds.
(speculation from here:) I suspect that the non-responsiveness is caused by something out there in the fediverse, _maybe_ caused by the "thread completion" feature other ostatus implementations seem to have.
Example: I create a post, but I don't have any subscribers on most of the thousands of mastodon instances, so my post would normally not federate to there. Then one well-known user (with followers on all the 1000 instances) replies to my post. Consequence: The 1000 mastodon instances might ask quitter.no (all at the same time) "hey quitter.no, give us that first post to which that well-known user replied to"
Die Digitalcourage-Hochschulgruppe schreibt im Fediverse:
„Panikmache und Falschinfos statt Datenschutz und Informationssicherheit: Ein Offizier der Bundeswehr erzählt vom Darkweb. Unsere Stellungnahme zum Vortrag vom 14. Mai an der Uni Bielefeld:“
@Digitalcourage hatte dieser Tage den Film "Das #Microsoft Dilemma" empfohlen. Nachdem ich mir den jetzt angesehen habe möchte ich das nochmal pushen.
Der Film schneidet das Problem von #closedSource Software an und zeigt ganz nebenbei noch ein paar grundsätztliche Schwächen unseres politischen Systems auf.
Er ist so aufbereitet, dass man ihn durchaus aus der Blase raus an Mutti oder den BWL-Kollegen verlinken kann.
@ouroboros gibts eigentlich bei Youtube schon Clips vom Geburtsvorgang? Kommt bestimmt auch noch. Und dann können die Zuschauer up- und downvoten und kommentieren.
is anyone working on saving or archiving any of the content from gnusocial.de? they are deleting everything in a few days. its one of the first large instances of the #fediverse and I am sure there is many important discussions that have taken place there that it would be historically relevant to keep. @textfiles ?
@patrickbreyer ich komme – mit etwas Verspätung – zurück auf die Frage nach GNUsocial-Instanzen, die keine IP-Adressen loggen. Leider kann ich dir jetzt quitter.se und quitter.no wegen anhaltender technischer Schwierigkeiten aktuell doch nicht empfehlen.
Jetzt kommt mir als nächstes gnusocial.ch in den Sinn...
Hallo @marcus: @patrickbreyer sucht wegen der Schließung von gnusocial.de eine neue Instanz, die keine IP-Adressen loggt. Ist das bei euch so?
Because: Some 3 weeks ago I told @knuthollund on #IRC about the issue and he said he would have a look. Soon after that the problem was completely gone and quitter.no worked flawlessly for at least the rest of the day! And moreover: quitter.se also worked flawlessly! But later it turned out @knuthollund didn't actually do anything at that point, not even restart anything.)
So I think that something on the fediverse network must have changed within those couple of hours. I think the cause of the problem is somewhere in the internet and that cause did stop its weird behaviour against quitter.se and quitter.no just for a while. Too bad the problem came back the next day.
I think one could check the logs of the webserver or use wireshark to look out for weird/excessive traffic. !gnusocial
@lakwnikos Regarding the "not-responding database server every few minutes" phenomenon I really wonder what the cause of it could be.
1) I think only quitter.se and quitter.no are affected. I did some research and asked on IRC but couldn't find any other instance with this problem yet
@knuthollund @lakwnikos Besides: as a user I can still live with issues like this. GNUsocial is a grassroot network for me and I like instances run by common people I like, even if there are some issues occasionally. It doesn't need to be "professional"! @hannes