- 7.1 Distributed Systems Operations
- 7.2 Service Life Cycle
- 7.3 Organizing Strategy for Operational Teams
- 7.4 Virtual Office
- 7.5 Summary
- Exercises
This chapter is from the book
This chapter is from the book
This chapter is from the book
7.3 Organizing Strategy for Operational Teams
An operational team needs to get work done. Therefore teams need a strategy that assures that all incoming work is received, scheduled, and completed. Broadly speaking, there are three sources of operational work and these work items fall into three categories. To understand how to best organize a team, first you must understand these sources and categories.
The three sources of work are life-cycle management, interacting with stake-holders, and process improvement and automation. Life-cycle management is the operational work involved in running the service. Interacting with stakeholders refers to both maintaining the relationship with people who use and depend on the service, and prioritizing and fulfilling their requests. Process improvement and automation is work inspired by the business desire for continuous improvement.
No matter the source, this work tends to fall into one of these three broad categories:
- Emergency Issues: Outages, and issues that indicate a pending outage that can be prevented, and emergency requests from other teams. Usually initiated by an alert sent by the monitoring system via SMS or pager. (Chapter 14)
- Normal Requests: Process work (repeatable processes that have not yet been automated), non-urgent trouble reports, informational questions, and initial consulting that results in larger projects. Usually initiated by a request ticket system. (Section 14.1.3)
- Project Work: Small and large projects that evolve the system. Managed with whatever project management style the team selects. (Section 12.4.2)
To assure that all sources and categories of work receive attention, we recommend this simple organizing principle: people should always be working on projects, with exceptions made to assure that emergency issues receive immediate attention and non-project customer requests are triaged and worked in a timely manner.
More specifically, at any given moment, the highest priority for one person on the team should be responding to emergencies, the highest priority for one other person on the team should be responding to normal requests, and the rest of the team should be focused on project work.
This is counter to the way operations teams often work: everyone running from emergency to emergency with no time for project work. If there is no effort dedicated to improving the situation, the team will simply run from emergency to emergency until they are burned out.
Major improvements come from project work. Project work requires concentration and focus. If you are constantly being interrupted with emergency issues and requests, you will not be able to get projects done. If an entire team is focused on emergencies and requests, nobody is working on projects.
It can be tempting to organize an operations team into three subteams, each focusing on one source of work or one category of work. Either of these approaches will create silos of responsibility. Process improvement is best done by the people involved in the process, not by observers.
To implement our recommended strategy, all members of the team focus on project work as their main priority. However, team members take turns being responsible for emergency issues as they arise. This responsibility is called oncall. Likewise, team members take turns being responsible for normal requests from other teams. This responsibility is called ticket duty.
It is common that oncall duty and ticket duty are scheduled in a rotation. For example, a team of eight people may use an eight-week cycle. Each person is assigned a week where he or she is on call: expected to respond to alerts, spending any remaining time on projects. Each person is also assigned a different week where he or she is on ticket duty: expected to focus on triaging and responding to request tickets first, working on other projects only if there is remaining time. This gives team members six weeks out of the cycle that can be focused on project work.
Limiting each rotation to a specific person makes for smoother handoffs to the next shift. In such a case, there are two people doing the handoff rather than a large operations team meeting. If more than 25 percent of a team needs to be dedicated to ticket duty and oncall, there is a serious problem with firefighting and a lack of automation.
The team manager should be part of the operational rotation. This practice ensures the manager is aware of the operational load and firefighting that goes on. It also ensures that nontechnical managers don’t accidentally get hired into the operations organization.
Teams may combine oncall and ticket duty into one position if the amount of work in those categories is sufficiently small. Some teams may need to designate multiple people to fill each role.
Project work is best done in small teams. Solo projects can damage a team by making members feel disconnected or by permitting individuals to work without constructive feedback. Designs are better with at least some peer review. Without feedback, members may end up working on projects they feel are important but have marginal benefit. Conversely, large teams often get stalled by lack of consensus. In their case, focusing on shipping quickly overcomes many of these problems. It helps by making progress visible to the project members, the wider team, and management. Course corrections are easier to make when feedback is frequent.
The Agile methodology, discussed in Section 8.6, is an effective way to organize project work.
7.3.1 Team Member Day Types
Now that we have established an organizing principle for the team’s work, each team member can organize his or her work based on what kind of day it is: a project-focused day, an oncall day, or a ticket duty day.
Project-Focused Days
Most days should be project days for operational staff. Specifically, most days should be spent developing software that automates or optimizes aspects of the team’s responsibilities. Non-software projects include shepherding a new launch or working with stakeholders on requirements for future releases.
Organizing the work of a team through a single bug tracking system has the benefit of reducing time spent checking different systems for status. Bug tracking systems provide an easy way for people to prioritize and track their work. On a typical project day the staff member starts by checking the bug tracking system to review the bugs assigned to him or her, or possibly to review unassigned issues of higher priority the team member might need to take on.
Software development in operations tends to mirror the Agile methodology: rather than making large, sudden changes, many small projects evolve the system over time. Chapter 12 will discuss automation and software engineering topics in more detail.
Projects that do not involve software development may involve technical work. Moving a service to a new datacenter is highly technical work that cannot be automated because it happens infrequently.
Operations staff tend not to physically touch hardware not just because of the heavy use of virtual machines, but also because even physical machines are located in datacenters that are located far away. Datacenter technicians act as remote hands, applying physical changes when needed.
Oncall Days
Oncall days are spent working on projects until an alert is received, usually by SMS, text message, or pager.
Once an alert is received, the issue is worked until it is resolved. Often there are multiple solutions to a problem, usually including one that will fix the problem quickly but temporarily and others that are long-term fixes. Generally the quick fix is employed because returning the service to normal operating parameters is paramount.
Once the alert is resolved, a number of other tasks should always be done. The alert should be categorized and annotated in some form of electronic alert journal so that trends may be discovered. If a quick fix was employed, a bug should be filed requesting a longer-term fix. The oncall person may take some time to update the playbook entry for this alert, thereby building organizational memory. If there was a user-visible outage or an SLA violation, a postmortem report should be written. An investigation should be conducted to ascertain the root cause of the problem. Writing a postmortem report, filing bugs, and root causes identification are all ways that we raise the visibility of issues so that they get attention. Otherwise, we will continually muddle through ad hoc workarounds and nothing will ever get better. Postmortem reports (possibly redacted for technical content) can be shared with the user community to build confidence in the service.
The benefit of having a specific person assigned to oncall duty at any given time is that it enables the rest of the team to remain focused on project work. Studies have found that the key to software developer productivity is to have long periods of uninterrupted time. That said, if a major crisis appears, the oncall person will pull people away from their projects to assist.
If oncall shifts are too long, the oncall person will be overloaded with follow-up work. If the shifts are too close together, there will not be time to complete the follow-up work. Many great ideas for new projects and improvements are first imagined while servicing alerts. Between oncall shifts people should have enough time to pursue such projects.
Chapter 14 will discuss oncall in greater detail.
Ticket Duty Days
Ticket duty days are spent working on requests from customers. Here the customers are the internal users of the service, such as other service teams that use your service’s API. These are not tickets from external users. Those items should be handled by customer support representatives.
While oncall is expected to have very fast reaction time, tickets generally have an expected response time measured in days.
Typical tickets may consist of questions about the service, which can lead to some consulting on how to use the service. They may also be requests for activation of a service, reports of problems or difficulties people are experiencing, and so forth. Sometimes tickets are created by automated systems. For example, a monitoring system may detect a situation that is not so urgent that it needs immediate response and may open a ticket instead.
Some long-running tickets left from the previous shift may need follow-up. Often there is a policy that if we are waiting for a reply from the customer, every three days the customer will be politely “poked” to make sure the issue is not forgotten. If the customer is waiting for follow-up from us, there may be a policy that urgent tickets will have a status update posted daily, with longer stretches of time for other priorities.
If a ticket will not be completed by the end of a shift, its status should be included in the shift report so that the next person can pick up where the previous person left off.
By dedicating a person to ticket duty, that individual can be more focused while responding to tickets. All tickets can be triaged and prioritized. There is more time to categorize tickets so that trends can be spotted. Efficiencies can be realized by batching up similar tickets to be done in a row. More importantly, by dedicating a person to tickets, that individual should have time to go deeper into each ticket: to update documentation and playbooks along the way, to deep-dive into bugs rather than find superficial workarounds, to fix complex broken processes. Ticket duty should not be a chore, but rather should be part of the strategy to reduce the overall work faced by the team.
Every operations team should have a goal of eliminating the need for people to open tickets with them, similar to how there should always be a goal to automate manual processes. A ticket requesting information is an indication that documentation should be improved. It is best to respond to the question by adding the requested information to the service’s FAQ or other user documentation and then directing the user to that document. Requests for service activation, allocations, or configuration changes indicate an opportunity to create a web-based portal or API to make such requests obsolete. Any ticket created by an automated system should have a corresponding playbook entry that explains how to process it, with a link to the bug ID requesting that the automation be improved to eliminate the need to open such tickets.
At the end of oncall and ticket duty shifts, it is common for the person to email out a shift report to the entire team. This report should mention any trends noticed and any advice or status information to be passed on to the next person. The oncall end-of-shift report should also include a log of which alerts were received and what was done in response.
When you are oncall or doing ticket duty, that is your main project. Other project work that is accomplished, if any, is a bonus. Management should not expect other projects to get done, nor should people be penalized for having the proper focus. When people end their oncall or ticket duty time, they should not complain that they weren’t able to get any project work done; their project, so to speak, was ticket duty.
7.3.2 Other Strategies
There are many other ways to organize the work of a team. The team can rotate though projects focused on a particular goal or subsystem, it can focus on reducing toil, or special days can be set aside for reducing technical debt.
Focus or Theme
One can pick a category of issues to focus on for a month or two, changing themes periodically or when the current theme is complete. For example, at the start of a theme, a number of security-related issues can be selected and everyone commit to focusing on them until they are complete. Once these items are complete, the next theme begins. Some common themes include monitoring, a particular service or subservice, or automating a particular task.
If the team cohesion was low, this can help everyone feel as if they are working as a team again. It can also enhance productivity: if everyone has familiarized themselves with the same part of the code base, everyone can do a better job of helping each other.
Introducing a theme can also provide a certain amount of motivation. If the team is looking forward to the next theme (because it is more interesting, novel, or fun), they will be motivated to meet the goals of the current theme so they can start the next one.
Toil Reduction
Toil is manual work that is particularly exhausting. If a team calculates the number of hours spent on toil versus normal project work, that ratio should be as low as possible. Management may set a threshold such that if it goes above 50 percent, the team pauses all new features and works to solve the big problems that are the source of so much toil. (See Section 12.4.2.)
Fix-It Days
A day (or series of days) can be set aside to reduce technical debt. Technical debt is the accumulation of small unfinished amounts of work. By themselves, these bits and pieces are not urgent, but the accumulation of them starts to become a problem. For example, a Documentation Fix-It Day would involve everyone stopping all other work to focus on bugs related to documentation that needs to be improved. Alternatively, a Fix-It Week might be declared to focus on bringing all monitoring configurations up to a particular standard.
Often teams turn fix-its into a game. For example, at the start a list of tasks (or bugs) is published. Prizes are given out to the people who resolve the most bugs. If done company-wide, teams may receive T-shirts for participating and/or prizes for completing the most tasks.