[CR] [gprd] Send 50% of main traffic to HAProxy 2.x

Production Change

Change Summary

In this change, all 39 new haproxy-main-* nodes will be added to the external Google Cloud load balancer along the existing 39 old fe-* nodes.

The reason I decided to add all the new nodes in one step and split the external traffic 50-50 between the old and new HAProxy nodes is based on previous CRs in which I tried to add only a few haproxy-main-* nodes. When we add a small number of nodes (i.e. 6), we see that roughly 50% of the total traffic comes through these new nodes! This quickly saturates the new nodes causing a number of issues. The rate of errors increases, the average response time increases, the queue length grows, the healtchecks start to fail, the logging buffers overflow, etc.

I have analyzed the last unsuccessful CR and the resulting incidents (1, 2) thoroughly. As a result, I have prepared all the metrics and dashboards to monitor the critical signals early in advance, so I can revert this change before it causes a considerable impact on our error budget.

In this CR, I'm adding 15 more haproxy-main-* nodes to the external load balancer for the main which currently has 39 old fe-* and 24 haproxy-main-* new nodes. This will bring the total amount of traffic going to new 2.x nodes to 50%.

Change Details

1. Services Impacted - ServiceHAProxy
2. Change Technician - @miladx
3. Change Reviewer - @mchacon3
4. Time tracking - 30 minutes
5. Downtime Component - none

Detailed steps for the change

Change Steps - steps to take to execute the change

Estimated Time to Complete (mins) - 30

• Set label changein-progress /label ~change::in-progress
• Merge https://ops.gitlab.net/gitlab-com/gl-infra/config-mgmt/-/merge_requests/6682
  • Ensure the post-merge pipelines all succeed.
• Validate the change in gprd environment:
  • Inspect the gitlab-lb HTTP header in the response and make sure the new HAProxy nodes are involved.
    • curl -s -I https://gitlab.com/my-public-space
    • curl -s -I https://gitlab.com/my-public-space/public-repo
• Check out the metrics listed in the monitoring section.
• Run the end-to-end QA tests against the gprd environment.
  • Run a pipeline on master branch here.
    • https://ops.gitlab.net/gitlab-org/quality/production/-/pipelines/2326735
• Set label changecomplete /label ~change::complete

Rollback

Rollback steps - steps to be taken in the event of a need to rollback this change

Estimated Time to Complete (mins) - 10

• Quick Fix: Shooting down the new HAProxy nodes in the Google Cloud console will cause the GCP load balancers to NOT send any traffic to these nodes. As a result, we only send traffic to the old HAProxy nodes.
• Revert https://ops.gitlab.net/gitlab-com/gl-infra/config-mgmt/-/merge_requests/6682
• Set label changeaborted /label ~change::aborted

Monitoring

• Dashboards
  • HAProxy 2
  • Frontend Overview (legacy name)
  • Host Stats
  • 🚨 LoadBalancer SLI Error Ration
  • 🔎 Fluentd SLI Error Ratio
• Logs
  • 🔎 Fluentd Errors for HAProxy
  • 🔎 Gitaly PostUploadPack errors
  • BigQuery haproxy-main-* Logs

Key metrics to observe

• 🚨 Metric: Load Balancer Error SLI
  • Thanos Graph
  • We do NOT expect any major change in the current trends (it should be less or around 0.01).
  • ⚠️ Any increase, jump, or major change in the current values.
  • 
• Metric: Current number of sessions
  • Frontend Sessions
  • Backend Sessions
  • Server Sessions
  • We expect a decrease in the number of sessions on the old nodes and a proportional increase on the new nodes.
  • ⚠️ If we start to see a disproportional influx of sessions on the new nodes.
• Metric: The rate of the total number of responses
  • Frontend Total Responses
  • Backend Total Responses
  • Server Total Responses
  • We expect a decrease in the rate of responses on the old nodes and a proportional increase on the new nodes.
  • You may want to switch between absolute values and/or rate values.
  • ⚠️ Any sudden increase or jump in the rate of total responses.
  • 
• Metric: The rate of the number of response errors
  • Backend Response Errors
  • Server Response Errors
  • We expect a decrease in the rate of response errors on the old nodes and a proportional increase on the new nodes.
  • You may want to switch between absolute values and/or rate values.
  • ⚠️ Any sudden increase or jump in the rate of response errors.
  • 
• Metric: The average response time of the last 1024 successful connections
  • Backend Average Response Time
  • Server Average Response Time
  • ⚠️ Look for any abnormal pattern like the following.
  • 
• Metric: The average queue time for the last 1024 successful connections
  • Backend Average Queue Time
  • Server Average Queue Time
  • ⚠️ Look for any abnormal pattern like the following.
  • 
• Metric: Gitaly request and concurrency metrics
  • Thanos Graph
  • We do NOT expect any significant change in the current trends.
  • ⚠️ If we start to see an increase in the no. of ResourceExhausted gRPC code in the first graph or a constant increase in the new two graphs.

Change Reviewer checklist

C4 C3 C2 C1:

• Check if the following applies:
  • The scheduled day and time of execution of the change is appropriate.
  • The change plan is technically accurate.
  • The change plan includes estimated timing values based on previous testing.
  • The change plan includes a viable rollback plan.
  • The specified metrics/monitoring dashboards provide sufficient visibility for the change.

C2 C1:

• Check if the following applies:
  • The complexity of the plan is appropriate for the corresponding risk of the change. (i.e. the plan contains clear details).
  • The change plan includes success measures for all steps/milestones during the execution.
  • The change adequately minimizes risk within the environment/service.
  • The performance implications of executing the change are well-understood and documented.
  • The specified metrics/monitoring dashboards provide sufficient visibility for the change.
    • If not, is it possible (or necessary) to make changes to observability platforms for added visibility?
  • The change has a primary and secondary SRE with knowledge of the details available during the change window.
  • The change window has been agreed with Release Managers in advance of the change. If the change is planned for APAC hours, this issue has an agreed pre-change approval.
  • The labels blocks deployments and/or blocks feature-flags are applied as necessary.

Change Technician checklist

• Check if all items below are complete:
  • The change plan is technically accurate.
  • This Change Issue is linked to the appropriate Issue and/or Epic
  • Change has been tested in staging and results noted in a comment on this issue.
  • A dry-run has been conducted and results noted in a comment on this issue.
  • The change execution window respects the Production Change Lock periods.
  • For C1 and C2 change issues, the change event is added to the GitLab Production calendar.
  • For C1 and C2 change issues, the SRE on-call has been informed prior to change being rolled out. (In #production channel, mention @sre-oncall and this issue and await their acknowledgement.)
  • For C1 and C2 change issues, the SRE on-call provided approval with the eoc_approved label on the issue.
  • For C1 and C2 change issues, the Infrastructure Manager provided approval with the manager_approved label on the issue.
  • Release managers have been informed prior to any C1, C2, or blocks deployments change being rolled out. (In #production channel, mention @release-managers and this issue and await their acknowledgment.)
  • There are currently no active incidents that are severity1 or severity2
  • If the change involves doing maintenance on a database host, an appropriate silence targeting the host(s) should be added for the duration of the change.