Support process

In Beta!

We are currently working out our support process. The content on this page might change over time, and we welcome suggested changes and pull requests!

This section contains information that our team uses to triage, communicate, and resolve Support Requests that come from communities.

Currently, Engineering has the authority to design the system of roles and processes that defines “support at 2i2c”, and to advocate for the resources needed to fill those roles. This will naturally be done in collaboration with others in 2i2c, but Engineering leads the effort and has an outsized role in the decision-making.

Overview of the support process

Here’s a brief overview of our support process[1]:

• A Community Representative sends a request to support@2i2c.org.

• This is triaged by our Support Steward team.

  • They assess whether they can resolve it quickly, and potentially do so.

  • If they cannot resolve it, then we raise this support issue with our engineering team.

• If the issue is an Incident

  • We will designate an Incident Commander to orchestrate our efforts around the incident.

  • We will prioritize resolving it over everything else.

  • We will open a dedicated Incident issue to track its progress.

  • We will provide regular communication to the Community Representative as we investigate and work to resolve the issue.

  • We will notify them once the incident is resolved.

• If the issue is a Change Request or Guidance Request

  • We will incorporate these requests into our work planning pipeline

  • We will prioritize with our other work based on impact, complexity, and whether we have dedicated funding for this work.

  • For these types of support, each community has a Support Budget that defines the hours we can sustainably spend on non-incident support requests.

• When any support issue is resolved, we will communicate with the Community Representative to confirm with them.

Types of support requests

Below are a few key terms that describe how we categorize Support Requests.

Support Request

Support Requests

Any request that a community sends to us via support@2i2c.org. Support requests are generally un-planned and happen in response to a community needing assistance with something unexpected. There are a few main categories of support that we consider, each is described below.

Incident

Incidents

An event that significantly degrades the JupyterHub service. Support requests that are related to incidents should be prioritized over all other work items.

What counts as an incident? defines the kind of incidents we respond to via PagerDuty and consider immediate issues to be resolved.

We do not have a limit on the support time we provide related to incidents (as opposed to Change and Guidance requests, which have a Support Budget).

See also

See Incident response for more information.

Change Request

Change Requests

A request for a desired change to a hub’s infrastructure that is not related to an incident. For example:

• Changing the user’s software environment.

• Changing the resources available to users.

• Updating and deploying changes from upstream tools for a community.

• Making an improvement to open source tools to benefit a community.

Change Requests are generally non-urgent and should not be associated with significant diminished service. They are often things that communities could carry out themselves with the proper guidance and infrastructure setup. We aim to make our hubs as configurable as possible by the community so that we are not on the critical path for things like environment updates.

Guidance Request

Guidance Requests

A Support Request that is not tied directly to a change in infrastructure. Sometimes support requests are not tied to a specific change, but a desire to discuss and request guidance. In this case we may set up a meeting to discuss as a group, or have some back-and-forth via support channels.

Roles and team structure

Supporting a 2i2c hub is a collaborative process between 2i2c and the community we serve.

The Support Team is one of the main teams in our Managed JupyterHub Service Team.

This consists of three main roles: Support Stewards, Community Representatives, and Hub Administrators.

Support Steward

Support Stewards

A two-person team of Support Stewards work together to triage and communicate with all external support requests. Tenure on the support team is for four weeks. Every two weeks (generally at the sprint meeting), a Support Steward cycles off the support team, and a new team member joins the team. The support team rotates through the “Open Infrastructure Engineering Team”, in alphabetical order.

The primary responsibilities of the Support Stewards are:

• Ensure that we meet our Support Service Level Objectives.

• Carry out our support process.

• Act as primary points of contact with Community Representatives.

• Trigger an Incident Response if need be.

Common alternate terms: Customer Liason, External Liason, or Customer Support.

Community Representative

Community Representatives

See the service documentation.

Hub Administrator

Hub Administrators

See the service documentation.

Who are our Support Team?

The Support Steward role rotates through the below Support Team members.

Communication channels

External communication

We use these channels for communicating with external stakeholders like Community Representatives:

• support@2i2c.org is our point-of-contact for all support-related external communication.

• The 2i2c FreshDesk account is where we track all support requests and communication.

• the “Get Support” page provides guidance that communities may follow to get support.

FreshDesk

We use FreshDesk to manage all of our hub support questions and change requests. It is at the following URL:

2i2c.freshdesk.com

Any emails sent to support@2i2c.org will be routed to this FreshDesk account, and all team members should have access to it.

Canned responses allow us to pre-populate common responses for many kinds of support.

Our canned responses.

Internal communication

We have a few channels for communicating around support requests:

• Our FreshDesk account allows for internal team communication via the Add Note button. This can be useful for sharing quick internal updates.

• The Eng & Prod board that collects support-related issues identified via metadata type==support.

• (deprecated) Issues with the support label were where we tracked support requests related to Change Requests and Guidance Requests.

Process for support triage and resolution

Here is the process that we follow when triaging and resolving support requests.

Triage process

The goal of the triage phase is to understand the Support Request, decide if it is related to an incident, and choose the appropriate resolution pathway.

This process is carried out in an ongoing basis by the Support Stewards.

1. Monitor our support channels. We use FreshDesk for all support requests, and the Support Stewards should regularly keep an eye on this account for new requests. When a new support requests comes in, move to the next step.

2. Read and understand. Within one working day[2], read the support request and try to understand what action would resolve it.

3. Decide if there is an incident. Determine if a request meets the definition of an incident.

4. Categorize the Support Ticket in FreshDesk.

5. If an Incident, go to Incident response process.

6. If not an Incident, go to Non-incident response process.

Non-incident response process

The goal of the non-incident response process is to bring standardization to our support response. This simple workflow tries to battle the bias towards a reactive response whereas it is also bringing some common patterns so all of our non-incident support responses are cohesive and shared among our support stewards. The current iteration of the workflow states each step and who should be responsible/accountable for the specific step, plus some other clarifications.

When a new ticket lands in Freshdesk under the support group and it is not an incident, we aim to respond within 24 working hours with a suggested next action. The next steps should be followed when resolving a ticket:

1. Who: Support steward

   First 24h initial ticket evaluation. In the first 24h a support ticket was opened, you should do an initial evaluation of the ticket and ask the Community Representative about any additional information you may need.

2. Who: Support steward

   Spend 30 minutes trying to resolve. If you believe you can resolve the issue within 30 minutes, try resolving it yourself.

   1. If you resolve the issue, then jump to the “Confirm resolution” step 7.

   2. If you don’t believe you can resolve the issue (or you couldn’t) in 30 minutes, jump to the next step.

   Follow the guide at Initial timeboxed (30m) ticket resolution checklist to try and reach to a decision.

3. Who: Support Steward

   Open an engineering issue. If this is a Change Request or Guidance Request and/or you cannot resolve the issue within 30 minutes, then open a support issue for the team to discuss.

   Open a general issue that describes the request and any next steps we should take. It should be automatically added to the Eng & Prod board by the existing automation. Add the type: support and choose an impact level in the Eng & Prod board metadata (available for the very same issue, no need to visit the board itself!).

   If the issue has a critical impact (we defer that first evaluation to the support steward), an additional ping to the support Slack channel is needed to boost the signal.

   What does critical mean?

   We recognize there might be some support-related issues that do not count as incidents, but they need a quick resolution (inside the current sprint window) because they are impacting the execution of desired or existing workflows (degraded experience) for our communities. Examples of those sorts of issues (requests) are:

   • Image refs updates

   • Profile updates

   • User storage limitations

   • Grafana (and Prometheus) failures

   Additionally and depending on the nature AND context of the issue (request):

   • Access to specific buckets

   • Authentication and authorization updates

   The support steward should self-assign the critical issue and work on it immediately (this is now outside of the 30-minute timebox described in step 2).

   If the support stewards (both of them) do not have the capacity to resolve the critical issue (ie. working on another critical issue, being out of their working time, etc.), they should ping the Engineering Manager (or the delegated person) so they can secure resources to resolve that issue on the fly (see step 7 below).

   The support steward should not work on issues with impact lower than critical (unless they are assigned as part of the “planned” reactive work in the context of a running sprint (see step 6 below).

4. Who: Partnerships representative and the Engineering Manager (or respective delegates)

   Revisit the impact metadata. Once a week (at minimum) the support view in the Eng & Prod board should be revisited to validate the impact level on support-related issues. Currently, we allocate a 30-minute working session every Wednesday (open to everyone to participate) to perform such impact revision and further prioritization (“planned” reactive) every other week (see step 7 for more details).

5. Who: Support steward

   Add a reference/link to the created engineering issue inside the Freshdesk ticket. You can use an internal note or make it public when you communicate back to the Community Representative in step 6. Also, move the status of the ticket to the “Pending” state.

6. Who: Support steward

   Communicate status. Once we have an issue created to track the next steps, send a message to the Community Representative letting them know about the situation: after some initial investigation and no immediate fix, a follow-up issue was created that will be assigned in the future accordingly to the current prioritization. Also, let them know what the next steps will be. Here’s a template to help guide you:

   Hello { NAME }, thanks for sending us your request. We've concluded that
   this is not related to an outage or diminished service, so we'll incorporate
   this request into our team's work planning process. We'll send an update
   when we've got a plan for completing this request.
   

7. Who: Engineering Manager (currently assigning reactive work) or someone delegated by the Engineering Manager

   Prioritize the request. Any non-critical issue should wait to be included in our sprints (on Wednesdays, every other week) to be worked out as part of the “planned” reactive work. Follow the how to prioritize Change and Guidance Requests guide to decide how we should prioritize this request relative to the other work we need to do. We should be fully transparent about the support queue to our Community Representatives if they ping us for updates.

   If there is any critical issue, we could assign people on the fly (during the sprint) to resolve them, but we should minimize that behavior (it should be exceptional cases).

8. Who: Support steward

   Resolve the request. When some engineer is assigned to a support-related GH issue in the context of a sprint, we move ahead with the investigation/resolution for one (1) sprint. If we failed to find a fix during that time, we communicate back that state in the Freshdesk ticket and resolve it.

   Exceptional tickets might need more than one sprint. These tickets need to be explicitly approved as exceptions.

9. Who: Support steward

   Confirm resolution. Once we have resolved a support request, send a message to the Community Representative to confirm that we believe it is resolved. In FreshDesk, mark the incident as Resolved.

10. Who: Support steward

    Close the request. If the Community Representative confirms that their request has been fulfilled, consider this request closed. In FreshDesk, mark the incident as Closed.

Prioritizing non-incident support requests

For support requests that are not related to Incidents, we have more flexibility in when to do the work. We cannot sustainably prioritize all support requests over our other work, so here is a rough prioritization order to follow:

1. Requests that come with dedicated funding.

2. Requests that impact many communities.

3. Requests that align with our pre-existing goals.

4. Requests that align with an upstream open source project’s goals.

5. Requests from communities that haven’t used much Support Budget.

6. Requests that have a high-impact on a single community.

7. Our other work items

8. Requests from communities that have no more Support Budget.

Ask the PMs

We have a few roles that are particularly useful for triaging and prioritizing our work items.

• Our Project Manager should have visibility over everybody’s workload, and helps us prioritize work during sprints.

• Our Product and Community Lead should understand which work items will be most impactful across our communities. (we currently do not have anybody serving in this role, but will soon!)

When deciding how to prioritize a Change Request, ask for guidance from these two team members.

Support request key terms

The following are other important terms and ideas in the support process.

Support Budget

A fixed amount of time that we can spend providing support for each community that we serve. This helps us ensure that we can sustainably serve many communities. Any support request that is not tied to an Incident will draw from the support budget for that community. If a community requests support beyond their support budget, we may request extra funds to help cover our costs.

Note

We currently keep this term intentionally vague, and ask that communities are respectful of our time when making change requests. We are investigating the support budget that we should give to each community, and will update here when we have specific numbers in mind.

Here is a rough idea of the rationale we follow as we identify more specific numbers for support budget:

• Assume 173 working hours a month per engineer.

• Assume any given engineer should spend no more than 20% of their time (one day a week) on average dealing with support requests.

• That amounts to 34 hours a month on support requests.

• Then choose a support budget for each community based on the number of communities we think one engineer can sustainably serve. Tie this number back to our question “how many hubs should be able to sustain a single engineer?”. That ranges between 20 and 8 (depending on edu/research, and dedicated cluster).

Expected time off and downtime

There are some cases when we intentionally reduce our operations and support capacity. See Expected end-of-year time off for our broader policies and support commitments during this time.

Expected time off support policy

Below is our policy for support during expected time off:

• We monitor our support e-mail for major incidents, but do not guarantee a response for non-essential requests or questions.

• We will take steps to minimize harm and avoid catastrophic problems (e.g. incidents that drastically increase costs), but will not perform non-essential actions. We will assume there is no essential work happening on any of our hubs for all communities.

• If there is a catastrophic incident, we will take the minimal number of actions to reduce risk and damage to a reasonable level.

• For non-catastrophic incidents and general change requests, we will wait until after this period to resolve them.

Support process during expected time off

Here are the steps we take to set expectations for our team and for other organizations during expected time off:

• One month before the start of time off. Add footer content to our support@2i2c.org and hello@2i2c.org responses that communicates our intent. See Add a footer to all replies for instructions on doing this.

  Example text:

  Note: the 2i2c team will have expected reduced capacity from STARTDATE to ENDDATE. During this time, we will provide operational support for significant cloud incidents and outages, but not for non-essential questions or change requests. We will be less responsive than normal, and will return to answer your questions and resolve issues after the time off.

• During the time off. Add an auto-responder to our FreshDesk accounts with a message like the following:

  Note: the 2i2c team is operating at an expected reduced capacity until ENDDATE. We will provide operational support for significant cloud incidents and outages, but not for non-essential questions or change requests. We will be less responsive than normal, and will return to answer your questions and resolve issues after the time off.

Rotating team members during expected time off

Because team members continue to serve as support stewards during this time, we should take care to avoid the same person serving in this role across multiple periods of expected time off.

---

[1]

We had a lot of discussion around various support and operations systems to take as inspiration. This GitHub Issue has a lot of useful discussion, including a few write-ups of specific support systems to use as inspiration (example one example two)

[2]

We define a working day as a continuous 24 hour period of time from Monday through Friday. This is because our team and the communities we serve are distributed across many time zones, so there is no single “working day” for everyone.