2017-12-01
- Uptime Robot noticed that CGSpace went down
- The logs say “Timeout waiting for idle object”
- PostgreSQL activity says there are 115 connections currently
- The list of connections to XMLUI and REST API for today:
# cat /var/log/nginx/rest.log /var/log/nginx/rest.log.1 /var/log/nginx/access.log /var/log/nginx/access.log.1 /var/log/nginx/library-access.log /var/log/nginx/library-access.log.1 | grep -E "1/Dec/2017" | awk '{print $1}' | sort -n | uniq -c | sort -h | tail
763 2.86.122.76
907 207.46.13.94
1018 157.55.39.206
1021 157.55.39.235
1407 66.249.66.70
1411 104.196.152.243
1503 50.116.102.77
1805 66.249.66.90
4007 70.32.83.92
6061 45.5.184.196
- The number of DSpace sessions isn’t even that high:
$ cat /home/cgspace.cgiar.org/log/dspace.log.2017-12-01 | grep -o -E 'session_id=[A-Z0-9]{32}' | sort -n | uniq | wc -l
5815
- Connections in the last two hours:
# cat /var/log/nginx/rest.log /var/log/nginx/rest.log.1 /var/log/nginx/access.log /var/log/nginx/access.log.1 /var/log/nginx/library-access.log /var/log/nginx/library-access.log.1 | grep -E "1/Dec/2017:(09|10)" | awk '{print $1}' | sort -n | uniq -c | sort -h | tail
78 93.160.60.22
101 40.77.167.122
113 66.249.66.70
129 157.55.39.206
130 157.55.39.235
135 40.77.167.58
164 68.180.229.254
177 87.100.118.220
188 66.249.66.90
314 2.86.122.76
- What the fuck is going on?
- I’ve never seen this 2.86.122.76 before, it has made quite a few unique Tomcat sessions today:
$ grep 2.86.122.76 /home/cgspace.cgiar.org/log/dspace.log.2017-12-01 | grep -o -E 'session_id=[A-Z0-9]{32}' | sort -n | uniq | wc -l
822
- Appears to be some new bot:
2.86.122.76 - - [01/Dec/2017:09:02:53 +0000] "GET /handle/10568/78444?show=full HTTP/1.1" 200 29307 "-" "Mozilla/3.0 (compatible; Indy Library)"
- I restarted Tomcat and everything came back up
- I can add Indy Library to the Tomcat crawler session manager valve but it would be nice if I could simply remap the useragent in nginx
- I will also add ‘Drupal’ to the Tomcat crawler session manager valve because there are Drupals out there harvesting and they should be considered as bots
# cat /var/log/nginx/rest.log /var/log/nginx/rest.log.1 /var/log/nginx/access.log /var/log/nginx/access.log.1 /var/log/nginx/library-access.log /var/log/nginx/library-access.log.1 | grep -E "1/Dec/2017" | grep Drupal | awk '{print $1}' | sort -n | uniq -c | sort -h | tail
3 54.75.205.145
6 70.32.83.92
14 2a01:7e00::f03c:91ff:fe18:7396
46 2001:4b99:1:1:216:3eff:fe2c:dc6c
319 2001:4b99:1:1:216:3eff:fe76:205b
2017-12-03
- Linode alerted that CGSpace’s load was 327.5% from 6 to 8 AM again
2017-12-04
- Linode alerted that CGSpace’s load was 255.5% from 8 to 10 AM again
- I looked at the Munin stats on DSpace Test (linode02) again to see how the PostgreSQL tweaks from a few weeks ago were holding up:
- The results look fantastic! So the
random_page_cost tweak is massively important for informing the PostgreSQL scheduler that there is no “cost” to accessing random pages, as we’re on an SSD!
- I guess we could probably even reduce the PostgreSQL connections in DSpace / PostgreSQL after using this
- Run system updates on DSpace Test (linode02) and reboot it
- I’m going to enable the PostgreSQL
random_page_cost tweak on CGSpace
- For reference, here is the past month’s connections:
2017-12-05