Upgrade
An upgrade, and any change to a value you set at install, is a rerun of the install command. Don't run helm upgrade on the Meko release by itself. The release keeps the BYOC defaults it was installed with, so a newer chart's defaults, including security fixes, never apply, and values you passed at install can be blanked.
Move to a new release
-
Get the install command again from the Meko console. It names the current release's chart version and image tags.
-
Pull the new chart at the
CHART_VERSIONin the new command, as in Edit the database definition. Copy the Foundation file for yourSIZEfrom it, and copy yourREPLACEedits into it from the file you installed with. The install stops before it changes anything if the file's release values differ from the chart's, and lists each one. -
Fill in the header with the same values as before, and set
FOUNDATION_CRto the new file.Leave
AGENT_API_KEYas it is. The rerun keeps the key that is installed. -
Run it.
The rerun keeps the secrets the first install generated: the MCP internal secret and Langfuse's keys. It applies the new chart's BYOC defaults.
A rerun installs only what the header and EXTRA_VALUES pass. If you set anything on the release by hand with helm upgrade, list it first:
helm get values meko -n <namespace>
Move what you find into EXTRA_VALUES, or the rerun drops it.
Change the size
Set SIZE='prod' in the header and run the install command again. A proof-of-concept install grows in place: the database's tablet servers and masters restart one at a time and their disks grow, and the Meko pods are resized.
Before you rerun:
- Move the database nodes to the production size (16 vCPU / 64 GiB). A production tablet server doesn't fit on a proof-of-concept node. The first one to restart would stay pending and leave the database mid-update.
- Copy your edits into the production definition. Each size is its own file. Start from
meko-foundation-cr-prod.yamland copy in your region, zones, storage classes,nodeSelector, and management account. The install stops before it changes anything if those differ from the running database.
A volume can grow but not shrink, so you can't go from prod back to poc. If a rerun leaves SIZE at its default of poc on a production install, the install refuses before it changes anything.
Move to a production address
To go from the load balancer's host name to your own DNS name:
- Request the certificate and note its ARN.
- Set
PUBLIC_URLandACM_CERT_ARNin the header, and run the install command again. - Create the DNS record that points your name at the load balancer.
- Register the new redirect and post-logout URIs with your identity provider. The rerun prints them.
The rerun updates the knowledge-base bucket's CORS rule for the new address, and leaves any other rule on the bucket alone.
Turn on invite email
Invite emails are off. A shared datapack appears under the recipient's incoming shares, and nobody is emailed. To send invitations:
-
In Amazon SES, verify a sender address.
-
Add
ses:SendEmailto themeko-byoc-bedrockuser's policy. -
Add the following to
EXTRA_VALUESin the header and run the install command again:EXTRA_VALUES="--set-string apiServer.ses.fromAddress=<sender> --set-string apiServer.ses.region=<region>"
Rotate credentials
To rotate an AWS key, create a new access key for the IAM user, put it in the header, rerun the install command, and then delete the old key. To rotate the OpenAI key, change OPENAI_KEY and rerun.
To rotate the Quay, metrics or log credentials, apply the new license file from Meko support in the same namespace. The install never replaces the license's Secrets.
The collectors read the metrics password and the log key from their Secrets when they start, so restart them after you apply a new license:
kubectl -n <namespace> rollout restart deploy/meko-metrics-collector daemonset/meko-logs-collector
Pods pull images with the Quay credentials when they start, so a new Quay credential applies to the next pod that starts.