Developer Offshore logoDeveloper OffshorePowered by StealthAgentsScope a developer
Table of contents

Developer Offshore guide

Write a capacity note for a seasonal traffic peak

How an offshore pod prepares a capacity note for a seasonal peak: define the window, separate load from capacity, check dependencies, plan staffing, and record stop conditions.

Write a capacity note for a seasonal traffic peak

Key takeaways

  • Define the peak window and the expected load shape before discussing servers.
  • Separate demand estimates from the capacity of each dependency in the request path.
  • Test the slowest dependency and the queue behavior, not only the application tier.
  • Match staffing and freeze windows to the local hours when the peak lands.
  • Write stop conditions that tell the on-call engineer when to shed load or escalate.

Define the peak window

A capacity note starts with a calendar, not a server count. Name the dates, the local hours, and the business events that drive the peak. Include the time zone of the users, because a peak in the buyer's morning may land outside the offshore pod's staffed hours. The window should be specific enough that the team can schedule a freeze around it.

Describe the expected load shape in plain terms. Note whether traffic arrives as a steady climb, a sharp opening, or a series of daily waves. Add the events that could double the estimate, such as a marketing send or a partner promotion. The shape tells the team which controls matter, because a short spike and a long plateau stress different parts of the system.

Separate load from capacity

Load is what the system receives, while capacity is what each component can absorb before latency or errors rise. Keep the two in separate sections so the note does not confuse a forecast with a limit. For load, record the request rate, the payload size, and the mix of read and write traffic. For capacity, record the measured limit and the headroom at that limit.

Use a measurement from a comparable period rather than a vendor ceiling. A database that claims a high connection count may still slow down well before that number under this query mix. Record the environment, the data volume, and the time of the test. A capacity number without those conditions is a marketing figure, not an operating limit. Repeat the test after any configuration change that touches connection pools, timeouts, or cache size.

Check the slowest dependency

Walk the request path and list every dependency that can add latency or fail. Include the payment provider, the identity service, the search cluster, and any third-party API with a quota. For each one, record the current limit and the behavior when that limit is reached. The slowest dependency usually sets the practical capacity of the whole path. Mark the dependencies owned by the buyer, because the offshore pod cannot change their limits alone.

Test the timeout and retry behavior under load rather than reading the configuration. A retry storm can turn a brief dependency slowdown into a sustained outage. Confirm that the client backs off, that the queue has a bound, and that the fallback path is honest about what it can show. Record the observed behavior so the peak plan reflects the system that exists.

Plan staffing around the peak

Match the support schedule to the hours when the peak is expected. If the offshore pod is asleep during the buyer's busiest window, the escalation path must be explicit and tested. Name the primary, the backup, and the buyer-side contact for each critical window. A schedule that only covers standard hours is not a peak plan. Publish the schedule in the support channel and test the escalation once before the window opens.

Decide which work pauses during the peak and which continues. A deployment freeze is common, but it should not stop a security patch or a fix for a known outage. Write the exception path for both. Also reserve one engineer for incident work instead of assigning the whole pod to feature delivery during the riskiest days. Tell the buyer which tickets will wait and which will continue, so expectations match the calendar.

Write the stop conditions

Give the on-call engineer observable thresholds for shedding load, disabling a non-critical feature, and escalating. Tie each threshold to a metric that is visible on the dashboard, such as error rate, queue depth, or checkout latency. State who can approve the next step and how quickly that approval is expected. A stop condition that requires a meeting will not be used during a spike.

Include the recovery path after a stop. If the team disables a feature, the note should say how to confirm the system is stable before re-enabling it. If the team sheds traffic, it should say which users are affected and how they are told. The note should also record the point at which the team stops investigating and focuses on communication.

Agree the communication plan

The capacity note should name who tells the buyer when a threshold is crossed. Write the first message in advance so the on-call engineer is not composing it during an incident. Include the observed metric, the user impact, the action taken, and the next update time. A short template removes several minutes of hesitation when the window is busy. Keep the template in the same runbook as the stop conditions so both are easy to find.

Decide which channel carries the update and who monitors it on each side. An offshore pod and a buyer in different time zones need a written record, not only a call. Confirm that the person who can approve load shedding is reachable during the peak. A communication path that ends at an unmonitored inbox is not a plan. Name a backup channel in case the primary tool is unavailable during the peak.

Review after the peak

Compare the observed load and latency with the note's estimate during the same window. Record where the estimate was wrong and which dependency became the constraint first. Add the incident links and the support volume for the period. This comparison is more useful than a general retrospective because it tests a specific forecast against a specific result. Keep the dashboard link and the query beside the comparison so the next review does not rebuild them.

Turn each gap into one follow-up with an owner and a date. A capacity note that is never compared with reality becomes a document the team writes because the process asks for it. Keep the completed note beside the next year's draft so the estimate improves. The pattern across two or three peaks is the real capacity knowledge. If the same gap appears in two peaks, treat it as a design issue rather than another note to write.

Use the assessment in your hiring plan

Questions about assessing Philippine developers

Who writes the capacity note?

The offshore engineer who understands the request path drafts it, and the buyer-side platform owner confirms the business forecast and the acceptable risk. Both should review the stop conditions.

How far ahead should the note be written?

Far enough to book tests, change freezes, and staff cover, usually four to six weeks before the peak. A note written the week before can still guide the response but cannot change capacity.

What if the forecast is uncertain?

Record a low, expected, and high case, and state which one the plan is built for. The stop conditions should be tied to observed load rather than to the forecast.

Sources

  1. Google SRE Book, Handling Overload
  2. NIST SP 800-61 Rev. 2, Computer Security Incident Handling Guide
  3. CISA Secure by Design

International Labour Organization guidance on remote work arrangements reinforces why remote role briefs should document expectations, communication rhythms, and accountable handoffs.