Cloud maintenance

Owned by Adam Monsen
Last updated: Sept 20, 2011 by Lindsey Livingston (Unlicensed)Version comment

Setup local machine

One time setup to start working with cloud. There are two major tools that need to be setup, the amazon ec2/rds tools and the chef command line client. This guide assumes Ubuntu desktop, lucid release (10.04). Later versions of Ubuntu might work as well. 

EC2/RDS command line tools

chef

Chef provides configuration management of machines, starts/stops services when configuration changes, etc.

Setup chef locally, create account, connect with your chef server. We've been using "Opscode console" (they host our chef server).

  • http://wiki.opscode.com/display/chef/Package+Installation+on+Debian+and+Ubuntu** just install "chef" package, chef-server not needed since we use opscode's
    • http://help.opscode.com/kb/start/2-setting-up-your-user-environment (scroll down to section Create your Chef repository)
    • organization key is in vault directory as mifos-validator.pem.cpt (same password as the password vault)
    • when setting up your chef env if you lost your client key or want to generate a new one go to this (replace with your username): http://community.opscode.com/users/YOUR_USERNAME/ (be sure to login again if you don't see the "get private key" link)
    • create ~/.chef/knife.rb. Here's a template: 
      # Replace USERNAME, ORGANIZATION with yours
current_dir = File.dirname(__FILE__)
log_level                :info
log_location             STDOUT
node_name                "USERNAME"
client_key               "#{current_dir}/USERNAME.pem"
validation_client_name   "ORGANIZATION-validator"
validation_key           "#{current_dir}/ORGANIZATION-validator.pem"
chef_server_url          "https://api.opscode.com/organizations/ORGANIZATION"
cache_type               'BasicFile'
cache_options( :path => "#{ENV['HOME']}/.chef/checksums" )
# Customize as necessary. Mifos cookbooks are in the cloud git
# repository, under chef/cookbooks. Multiple paths are allowed.
cookbook_path            ["#{ENV['HOME']}/git/mifos-cloud/chef/cookbooks"]
    • Copy the keys and knife configuration you downloaded earlier into ~/.chef: 
      $ mkdir -p ~/.chef
$ cp USERNAME.pem ~/.chef
$ cp ORGANIZATION-validator.pem ~/.chef
$ cp knife.rb ~/.chef
    • verify your connectivity:
knife node list

you should see a list of nodes that are currently managed by chef

Knife hints

Cookbooks

List cookbooks that chef server knows about

knife cookbook list

Updating a cookbook

Cookbooks are stored in git in the chef directory of the cloud repository.  If you want to update a cookbook, UPDATE/COMMIT/PUSH IN GIT FIRST AND BUMP VERSION NUMBER, before sending it to chef server.  Here is a step by step instructions

$ mkdir -p ~/git/mifos-cloud
$ git clone git://mifos.git.sourceforge.net/gitroot/mifos/cloud ~/git/mifos-cloud
$ cd ~/git/mifos-cloud
# update version before doing anything: vi chef/cookbooks/<cookbook>/metadata.rb
$ make changes
$ knife cookbook upload <cookbook you changed>
$ git add/commit/push

How to get the AMI of every node

knife search node "ec2:[* TO *]" -a ec2.ami_id

How to get both AMI and hostname of every node

knife exec -E 'nodes.all {|n| if n.name =~ /^i-/ then printf("%-45s\t%s\n" % [n.override.tomcat.vhost,n.ec2.ami_id]) end }'|sort

(Thanks, opscode!)

How to get whats installed for each MFI

You will need to install amazon-ec2 gem e.g. sudo gem install amazon-ec2 and knife should be working. Invoke maint/state.rb like so:

knife exec state.rb

How to change Pentaho to run reports and ETL against an RDS replica

If you wish to use the RDS instance for Mifos and Pentaho, ignore this section.

1. Set up RDS replica.

2.

knife role edit mifos_MFI

3. Edit override_attributes.pentaho.mifos_database_replica_host (optionally, adding this setting). "null" means fall back to override_attributes.mifos.database_host, and is the same as omitting override_attributes.pentaho.mifos_database_replica_host (see cookbooks/pentaho/recipes/default.rb in the cloud repo for details).

NOTE: nothing maintains Pentaho's database (ex: "MFISHORTNAME_prod_hib"), so "SourceDB" must be changed manually here.

NOTE: data sources in BIRT reports must be maintained manually, separately.

NOTE: data sources in Jasper Servers must also be maintained manually, separately.

Starting a new mifos/pentaho instance

  • if converting an mfi from old infrastruture
    • Stop mifos instance
    • Dump database
    • Copy uploads/config in MIFOS_CONF
  • Create security group in EC2 (AWS UI)
  • SSH "gateways" setup/info
    • allows us to limit points of entry for our hosted machines
    • note hosts below in ec2-authorize commands... the gateways are currently birch.mifos.org (the whole Seattle GTC, actually) and cloudboss.mifos.org.
    • add to your .ssh/config (substituting MFINAME for something meaningful): 
      Host *MFINAME.mifos.org
   ProxyCommand ssh birch.mifos.org exec /bin/nc %h %p
  • one-time setup for EC2 physical firewall
    • manually change EC2_ACCOUNT_NUMBER with the 12-or-so digit number fetched from the AWS console
    • SSH via gateways only - also images will constantly change so you WILL see the "WARNING: REMOTE HOST IDENTIFICATION HAS CHANGED" often. You can verify the key if you want to with ec2-get-console-output or via the ec2 web ui.
    • 18980-18981 is for monitoring JMX over RMI via OpenNMS
    • Create security group first via web ui or cli
        
      #!/bin/bash
# login through web console and create new security group, e.g. green-grameen
set -ex
SEC_GROUPS="digamber light-microfinance rise keef"
EC2_ACCOUNT_NUMBER=000000000000

for SEC_GROUP in ${SEC_GROUPS}
do
ec2-authorize -P tcp -p 22-22 -s 75.149.167.24/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 22-22 -s 10.252.50.116/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 22-22 -s 184.72.240.48/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 443-443 -s 0.0.0.0/0 ${SEC_GROUP}
ec2-authorize -P tcp -p 80-80 -s 0.0.0.0/0 ${SEC_GROUP}
ec2-authorize -P tcp -p 18980-18981 -s 10.252.50.116/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 18980-18981 -s 184.72.240.48/32 ${SEC_GROUP}
ec2-authorize -P udp -p 161-161 -s 10.252.50.116/32 ${SEC_GROUP}
ec2-authorize -P udp -p 161-161 -s 184.72.240.48/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 161-161 -s 10.252.50.116/32 ${SEC_GROUP}
ec2-authorize -P tcp -p 161-161 -s 184.72.240.48/32 ${SEC_GROUP}
ec2-authorize -P icmp -t -1:-1 -s 10.252.50.116/32 ${SEC_GROUP}
ec2-authorize -P icmp -t -1:-1 -s 184.72.240.48/32 ${SEC_GROUP}
ec2-authorize -o ${SEC_GROUP} -u ${EC2_ACCOUNT_NUMBER} ldap
done
  • Create RDS security group
    • Authorize EC2 security group for MFI
    • Authorize default EC2 security group temporarily to make importing existing database more straightforward, remove after importing
  • Create RDS instances as m1.small initially
    • v5.1.50
    • enable auto minor version upgrade
    • allocate 10GB
    • use MFI long name for MySQL instance (ie: "rise", "secdep")
    • initial user/pass can be anything simple, this will be changed later
    • leave Database Name blank
    • Db Parameter Group: "mifoscloud"
    • backup retention period: 8 days (best for PITR/binlogs)
    • backup window: 1600-1700 UTC (good for India/Philippines/Africa)
    • maintenance windows Saturday 1700-1800 UTC
    • example: 
      Engine: mysql
Engine Version: 5.1.50
Auto Minor Ver. Upgrade: Yes
DB Instance Class: db.m1.small
Multi-AZ Deployment: Yes
Allocated Storage: 10
DB Instance Identifier: rise
Master User Name: mifos
Master User Password: mifos
Database Name:
Database Port: 3306
Availability Zone: Using a Multi-AZ Deployment disables this preference.
DB Parameter Group: mifoscloud
DB Security Group(s):rise
Backup Retention Period: 8
Backup Window: 16:00-17:00
Maintenance Window: Saturday 17:00-Saturday 18:00
  • Create chef roles, base + test + prod + optional MFI specific recipe
    • look at an existing role
      • knife role show mifos_digamber
    • create a new role
      • knife role create mifos_rise
      • knife role create mifos_rise_test
        (look at mifos_digamber_prod, mifos_digamber_test for examples)
  • Create 2 EBS volumes 1 for test and prod each (storing uploads)
    for testing: 
    ec2-create-volume --snapshot snap-5abd2f36 -s 1 -z us-east-1d
ec2-create-tags -t Name=testing-digamber <vol-id>
    for prod: 
    ec2-create-volume --snapshot snap-5abd2f36 -s 1 -z us-east-1a
ec2-create-tags -t Name=digamber <vol-id>
  • Update DNS if required, this is typically required if this is a new mfi.
    • Allocate elastic ip
    • nslookup ip and get hostname
    • create CNAME in network solutions management console (use CNAME instead of A record so inter/intra-AZ data transfer will be charged at lower rate)
  • Create EC2 instances
    • get ami-id from hudson job: https://ci.mifos.org/hudson/view/cloud/job/cloud-mifos-image
    • latest ami-id for 2.0.2:ami-8a8d7fe3 (see at end: https://ci.mifos.org/hudson/view/cloud/job/cloud-mifos-image/44/console) 
      ec2-run-instances ami-8a8d7fe3 --instance-type m1.small -z us-east-1d -d '{ "run_list":  ["role[ldapclient]", "role[base]" ] }' --disable-api-termination -g rise
ec2-create-tags -t Name=testing-rise.mifos.org -t Service=Mifos INSTANCE_ID
    • on boot, the node will add itself to the chef server (see rc.local, imaging/create_image.py, cloud source code)
    • make sure you can login via SSH. If not, fetch console output (this is something you may have to do from time to time): 
      ec2-get-console-output INSTANCE_ID
    • attach ec2 volume 
      ec2-attach-volume -i INSTANCE_ID -d /dev/sdc1 vol-ee7e2386
    • knife node edit INSTANCE_ID.mifos.org, add "role[mifos_rise_test]" to run list section or do through http://manage.opscode.com
    • log into box and run sudo chef-client to see change immediately or wait 30 minutes or so
  • set up Mifos and Pentaho databases 
    sudo /etc/pentaho/system/mifos_pentaho_init.sql -u mifos -pmifos
    • change mifos password via AWS Web UI (Modify RDS instance, put a password generated with, for example, apg, in the "Master User Password" field)
  • add backup jobs to BackupPC
    • when adding new backup host, use the NEWHOST=COPYHOST syntax mentioned on the "edit hosts" page
  • add monitoring of box to OpenNMS

Monitoring systems

Disaster recovery

Database

Most of persistent data is stored in RDS.  This implies the data is highly available as it is replicated synchronously in two availability zones.  However, it is certainly not impossible to lose an entire region e.g. due to natural disaster etc.  In addition to relying on multi-AZ functionality we also save and encrypt daily full mysqldump to cloudboss (in us-east-1b AZ) here: https://cloudboss.mifos.org/cloud   In the event of disaster you would need to download and decrypt the snapshot and create a new RDS instance and follow the instructions that apply when migrating an MFI from the old infrastructure.

Point in time recovery

RDS supports point in time recovery.  We configure each RDS instance to store up to 8 days of PITR logs.  However this feature in our trial runs can be really slow e.g. 9 hours or so to do a restore.  As an alternative you can use the mysqldump snapshots from above.

Front-end

Application server (Tomcat/Jetty). What clients hit.

In the event of AZ being unavailable or hardware failure.

            •  Identify which situation by checking if other nodes in the same AZ are available or not.
  • If hardware failure, simply launch a new instance with the appropriate AMI, add it to chef config, remap elastic ip, mount volumes, etc.
  • If an entire AZ is down:
  • create new volumes in alternate AZ, retrieve from backuppc the uploads, custom reports etc.
    • only use "tar download" restore method, and only of /etc/mifos/uploads dir
    • download tar to local machine, then copy to remote host and untar as user "tomcat6"
  • relaunch each frontend into an alternate AZ, add to chef config, remap elastic ips, mount new volumes etc.

If you manually stop Mifos, for example, during a restore of /etc/mifos/uploads, Chef will automatically restart it. To temporarily disable this behavior, you can use: sudo service chef-client stop, then sudo service chef-client start when you're finished.

Statefiles

Statefiles are lists of specific versions of packages to be included in images. They are kept in the statefiles/ dir in the "cloud" git repo.

  • updated from ci periodically (right now * */3 * * *)
  • commited/tagged/pushed to "cloud" git repo at sf.net if there is a change (can monitor commit logs) to be notified of a change
  • tag has build number and job name

Image maintenance

When upgrading machines, be sure to schedule outages.

Upgrades for new Mifos versions

If a new point release of Mifos comes out (e.g. 2.1.9):

  • modify the AMI generation script for that Mifos release to use the new point release version. (We would modify imaging/mifos_2_1_bi_1_2.sh to get the 2.1.x Mifos version along with BI 1.2).
  • update the "mifosversion" variable in the script to be 2.1.9 (commit and push)
  • re-run the hudson job "cloud-mifos_2_1-bi_1_2-image" to create a new AMI with the updated Mifos war (the name of the new AMI will be in the console log output on of the hudson job).
  • follow the groovy script usage below using the new AMI generated in the previous step.

Upgrades for security/features

  • Statefiles with lists of latest packages are created periodically (see above).
  • An administrator must keep track of security releases in upstream Ubuntu packages.
  • A groovy script is available to move a customer from one image to the next (it could be an upgrade or just a security update), this script should only be used if the mfi deployment for the environment already exists, volumes created, elastic ips associated etc. It can be invoked like so:
groovy upgrade.groovy <mfi e.g. rise> <environment testing|prod> <ami id>

It will then prompt you to verify to continue.  You should check the AMI path, calculated hostnames, etc.  There will be roughly 5 minutes of downtime to change the image (as long as there is not a major mifos upgrade requiring database migrations).

This script requries groovy 1.7 or later. It also needs the following shell variables to be defined:

export AMAZON_ACCESS_KEY_ID=xyz
export AMAZON_SECRET_ACCESS_KEY=xyz

TODO: how to move a customer to a newer image, what adhoc/manual tests to perform after bouncing their servers, how/when to notify customer.s of the change(s)

When using the upgrade script use the "Long Name" from below for the MFI argument

Long Name

Short Name

secdep

sec

rise

ris

light-microfinance

lmf

digamber

dig

keef

kee

Image production

Image production ci jobs are manually kicked off since there is a cost associated with storing Amazon EC2 images. Fire off the cloud-mifos-image job on the ci server to create a new image.

LDAP

See also: /wiki/spaces/MIFOSADMIN/pages/8552788 (login to mifosforge.jira.com required).

Pre-requisites

  • Log into a machine already in ldap or ldap.mifos.org
  • Create ~/.ldaprc and copy the following lines into it if it is already not there:
TLS_CERT /etc/ssl/certs/ldap.crt
TLS_KEY /etc/ssl/private/ldap.key
  • The passphrase required when invoking the commands below can be found in vault.
  • Admin accounts are added to group id 10000 and will get sudo on all ldap clients, automation accounts e.g. backuppc accounts are added to group id 11000, "regular"/non-admin users are added to 12000

Adding a new user

You can use the following script to help generate the ldif formatted input:

#!/bin/sh

# ./addnewuser.sh johndoe John Doe <UID> johndoe@grameenfoundation.org <secret>

cat << EOF
dn: uid=$1,ou=people,dc=mifos,dc=org
objectClass: inetOrgPerson
objectClass: posixAccount
objectClass: shadowAccount
uid: $1
sn: $3
givenName: $2
cn: $2 $3
displayName: $2 $3
uidNumber: $4
gidNumber: 10000
gecos: $2 $3
loginShell: /bin/bash
homeDirectory: /home/$1
userPassword: $6
shadowExpire: -1
shadowFlag: 0
shadowWarning: 7
shadowMin: 8
shadowMax: 999999
shadowLastChange: 10877
mail: $5
EOF

then invoke it like:

./addnewuser.sh johndoe John Doe <UID> johndoe@grameenfoundation.org <secret> | ldapadd -x -W -D cn=admin,dc=mifos,dc=org -h ldap.mifos.org -ZZ

if the user "johndoe" exists it will error, however duplicating the UID will NOT generate an error so be sure to make sure that it does not exist. If the command completes successfully than you can log into a ldap client e.g. <mfi>.mifos.org and run getent passwd and you will see the newly added entry.

You will need to add an authorized_key for the user in the chef config:

knife data bag create authorized_keys <user_id>
{
  "id": "<user id>",
  "key": "ssh-rsa <key content>"
}

Deleting a user

If you make a mistake you can delete the entry with the following:

ldapdelete -x -W -D cn=admin,dc=mifos,dc=org -h ldap.mifos.org -ZZ 'uid=johndoe,ou=people,dc=mifos,dc=org'

Also, you should fill in their data bag with an invalid key e.g.:

knife data bag show authorized_keys jbrewster
{
  "id": "jbrewster",
  "key": "ssh-rsa DISABLED"
}

Searching

You can also search the ldap db with the following

ldapsearch -LLL -x -W -D cn=admin,dc=mifos,dc=org -h ldap.mifos.org -ZZ 'cn=*Jeff*'

The last argument 'cn=Jeff' can be adjusted accordingly based on which field you want to search on.

Resetting a password

The following script will generate the ldif formatted data to feed into ldapmodify

#!/bin/sh

# ./reset.sh johndoe <secret>

cat << EOF
dn: uid=$1,ou=people,dc=mifos,dc=org
changetype: modify
replace: userPassword
userPassword: $2
EOF

then you can invoke it like so:

./reset.sh johndoe <THE NEW PASSWORD> | sudo ldapmodify -x -W -D cn=admin,dc=mifos,dc=org -h ldap.mifos.org -ZZ

At this point you'll need to enter the LDAP master password (from the vault).

SECDEP Specific Configuration

Secdep MFI has one additional detail that is not managed via chef. This MFI has a jasper reporting server (reports.mifos.org) that has a reference to the mifos datasources. When changing database hosts, etc. you MUST verify that the jasper reports continue to work. To update a datasources:

1. login to http://reports.mifos.org and click on
2. click on the second drop down in the "Refine" section and select "Data Sources"
3. Right click on the "Secdep Prod Mifos DB" datasource and click "Edit"
4. Update the credentials to the new host.
5. If the database host is remote, it must use ssl in order to connect, this can be enforced with the following url params: verifyServerCertificate=true&useSSL=true&requireSSL=true
6. If after updating the credentials and clicking on "Test Connection" and the connection fails verification than verify with the mysql cli that you can connect e.g. mysql -h <host> -u <user> -p --ssl --ssl-ca=<path to ca cert pem> --ssl-verify-server-cert and verify you can connect that way. If it does succeed most likely the java truststore has not been updated with the ca cert. You can trust it with this command

/home/mifosadmin/jasper_oss/jasperserver-ce-3.7.0/java/bin/keytool -import -file ca-cert.pem -alias ca-cert -keystore mysql-ssl-ca-cert.ts -storepass <store pass>

7. Restart jasper if the trust store has been updated. 

sudo service secdepjasper restart

Updating Jasper reports

1. Log into jasper
2. Right click on report, click "Edit"
3. Click next to update JRXML files
4. Browse for reports here: http://mifos.git.sourceforge.net/git/gitweb.cgi?p=mifos/documents;a=tree;f=deployment/SECDEP/Jasper+Reports;h=f80156caba0e3d03900bda404662e99b96711bfd;hb=HEAD

Note: PCFC report needs to select each jrxml file

MySQL/RDS maintenance

Changing database hosts

If for some reason the database coordinates need to be changed than the following steps need to be taken:

  • Update role (usually) mifos_<MFI>
  • run chef-client on each host to see changes immediately
  • update uploaded reports in /etc/mifos/uploads/report with something like: sudo find /etc/mifos/uploads/reports -type f -exec sed -i -e 's/secdep-db.mifos.org/secdep.cz2a1vveusgo.us-east-1.rds.amazonaws.com/g' {} \;
  • update pentaho datasources in <mfi_shortname>_<environment>_hib e.g. sec_prod_hib with a query similar to:

for DestinationDB:

update DATASOURCE set URL = 'jdbc:mysql://<HOST>:3306/sec_prod_mifos_dw?useUnicode=true&characterEncoding=UTF-8' where NAME='DestinationDB';

and for SourceDB:

update DATASOURCE set URL = 'jdbc:mysql://<HOST>:3306/sec_prod_mifos?useUnicode=true&characterEncoding=UTF-8' where NAME='SourceDB';

remember to update for both prod and test environments!

Growing a database

Log into RDS console at: https://console.aws.amazon.com/rds/ right-click on instance and click "Modify" and adjust allocated storage appropriately. The database will be down while it resizes so plan accordingly when you decide to increase the size.

Changing master user password

Do this from the AWS console. Make sure you check the box next to "Apply Immediately", or you may have to wait some amount of time (maybe a few minutes) before your changes are applied.

ETL configuration

An ETL job copies data from the Mifos OLTP database to the data warehouse. See these files in the mifos-cloud repository:

  • chef/cookbooks/pentaho/recipes/default.rb
  • chef/cookbooks/pentaho/templates/default/data_integration.sh.erb

Adding a new package to be part of image

Update create_mifos_state.sh, add package to add to the PACKAGES variable

TODO

  • deploy/migration procedure (from old EC2 setup to new cloud images)
    • document
    • practice
  • new disaster recovery procedure
    • document
    • practice
  • document new system
    • architecture
    • persistent data
    • configuration
  • separate setup/initial instructions from specific maintenance procedures
    • don't need step: "clone chef repo"
  • commit cookbooks into the cloud repo
    • DONE 2011-03-14 by Adam
  • script to wrap ec2-run-instances
    • keep trying to attach volume until succeeds immediately after running "ec2-run-instances"
    • nevermind... changed procedure so mifos_X role isn't included
  • be able to generate both 2.0 and 2.1 images etc.
    • aka branches, parallel development so release maintenance and new development can occur (currently we just have one Mifos 2.0 image, cloud/master can't handle branched development. Statefile probably needs to be maintained outside of the "cloud" repo)
  • groovy script to reimage a box (for instance, for a security upgrade)
  • do a couple of test restores
  • we can't currently pin recipe/role versions in a role run_list
    • chef client crashes!

Trash

  • Clone chef repo 
    $ cd ~
$ git clone http://github.com/opscode/chef-repo.git
$ mkdir -p ~/chef-repo/.chef