search

input-output-hk / cardano-ops

NixOps deployment configuration for IOHK/Cardano devops
Apache License 2.0
65 stars 160 forks source link

readme

cardano-ops

NixOps deployment configuration for IOHK/Cardano devops.

For examples on how you can deploy your own testnet please refer to the REAME of the Shelley testnet example.

Explorer upgrades

The explorer is composed with a traefik frontend (explorer) and two backends (explorer-a and explorer-b). The backends are dedicated machines, unless globals.explorerBackendsInContainers = true; in which case the backends are packed together (as containers) with the traefik frontend.

We upgrade one backend after the other to achieve zero-downtime upgrades.

On testnet and mainnet, after the upgrade is completed, a snapshot of the cardano-db-sync database is done and uploaded to s3 buckets:

Process

  1. add pins for a set of new versions of explorer services (that work together), eg.:
niv add input-output-hk/cardano-db-sync -n cardano-db-sync-10 -b refs/tags/10.0.0
niv add input-output-hk/cardano-graphql -n cardano-graphql-next -b chore/cardano-db-sync-10-compat
niv add input-output-hk/cardano-explorer-app -n cardano-explorer-app-next -b chore/cardano-graphql-5.0.0-compat
  1. Create a set for those new versions in globals-default.nix

Set one of backend (b) to use this new set (explorer10), and keep only the untouched backend in the active set:

  explorerBackends = {
    # explorer-a is updated to use the new set:
    a = globals.explorer10;
    b = globals.explorer9;
  };

  # explorer-a is being upgraded: we remove it from traefik load-balencer:
  explorerActiveBackends = ["b"];

  # new set of versions (to be updated with final tags before mainnet release)
  explorer10 = {
    cardano-db-sync = sourcePaths.cardano-db-sync-10;
    cardano-explorer-app = sourcePaths.cardano-explorer-app-next;
    cardano-graphql = sourcePaths.cardano-graphql-next;
  };

Commit this change to new branch and deploy it (explorer on shelley-qa or explorer-a on staging/testnet).

If this is a major upgrade, database on explorer-a need to be deleted: systemctl stop postgresql.service && rm -rf /var/lib/postgresql/12 && systemctl start postgresql.service && systemctl restart cardano-db-sync.

  1. Take a snapshot on explorer-a:

First we need to wait until cardano-db-sync is fully synced. Then we modify topology file to include this bit:

explorer-a.services.cardano-db-sync.takeSnapshot = "once";
  1. Swith frontend to updated backend (explorer-a) and prepare upgrade of explorer-b:

Edit globals-default.nix so that explorer-a use the new version, and the traefik frontend use the new version on explorer-b.

  explorerBackends = {
    a = globals.explorer10;
    # we now update explorer-b:
    b = globals.explorer10;
  };

  # explorer-a is now fully synced and ready to serve requests:
  explorerActiveBackends = ["a"];

Deploy frontend:

$ nixops --include explorer

At this point please check explorer web ui and rollback this last change if there is any issue.

  1. Upgrade explorer-b using the snapshot
$ nixops ssh explorer-a -- ls /var/lib/cexplorer/*tgz
/var/lib/cexplorer/db-sync-snapshot-schema-10-block-5886057-x86_64.tgz

$ nixops scp --from explorer-a /var/lib/cexplorer/db-sync-snapshot-schema-10-block-5886057-x86_64.tgz ./
$ nixops scp --to explorer-b db-sync-snapshot-schema-10-block-5886057-x86_64.tgz /var/lib/cexplorer/

$ nixops deploy --include explorer-b

Then wait for explorer-b to be fully synced.

  1. Update frontend to use both backend

Edit globals-default.nix to activate both backends:

  explorerBackends = {
    a = globals.explorer10;
    b = globals.explorer10;
  };
  explorerActiveBackends = ["a" "b"];

Push this change to the branch and merge it to master.

$ nixops deploy --include explorer
  1. Upload snapshot to S3

On testnet:

source ../proposal-ui/static/proposal-ui-testnet.sh
./scripts/checksum-sign-upload.sh db-sync-snapshot-schema-10-block-2700107-x86_64.tgz updates-cardano-testnet cardano-db-sync

On mainnet:

source ../proposal-ui/static/proposal-ui-mainnet.sh
./scripts/checksum-sign-upload.sh db-sync-snapshot-schema-10-block-2700107-x86_64.tgz update-cardano-mainnet.iohk.io cardano-db-sync

Accessing Prometheus

It is possible to query Prometheus instances directly (rather than via Grafana using the Prometheus query language, for example

cardano_node_metrics_utxoSize_int{hostname="stk-a-1-IOG1-ip"}[5m]

For larger queries, replacing 5m (minutes) by 5d (days) the GUI is inconvenient and it is better to use a programming environment to submit an HTTP request and parse the response. One way to do this is to use Firefox as described here.

Using this may give you several possible HTTP requests:

Choose the one that corresponds to the required query and then copy as cURL and execute it at the command line. It should also be possible to use this in a programming language such as Python.

DB Sync Best Practices

The database used in Cardano DB Sync is a PostgreSQL database. Like all databases, it needs to be set up in the best way for the intended environment. For best performance, the db-sync process needs to run on the same machine as the PostgreSQL server and the PostgreSQL database. For more information, see the recommendation from the developers.