Looker API rate limit hit mid-dashboard-build: agent failed with error 4

Not specified by the original author.

TL;DR

Slow the agent down and stop it re-asking questions it already answered. Read the retry guidance in the 429 response where the Looker API 4.0 provides it, back off with jitter, batch or cache the read calls the build repeats, and checkpoint progress so a throttled run resumes instead of restarting and re-spending its call budget. Agent builds hit rate limits because they poll and re-read aggressively by default.

Looker API rate limit hit mid-dashboard-build: agent failed with error 429 too many requests

Use this when

• Run logs show 429 responses from the Looker API 4.0 partway through a build that starts cleanly.
• The failures cluster around polling, listing, or read back heavy steps.
• Several agent runs or integrations share the same API credentials.

Not for this skill when

• A single 429 appears once and the run recovers by itself. Note it, but no redesign is needed.
• The API returns a quota or billing error rather than a rate limit. That is an account limit, handled with the provider.

Steps

1. Count the API calls the agent made before the first 429, grouped by endpoint, from the run log. Verify: the heaviest endpoints are named with their call counts
2. Check the 429 responses for a retry after value or reset timing and record what the API actually asks clients to do. Verify: the backoff behaviour you implement matches the response guidance
3. Implement exponential backoff with jitter on 429 responses, with a maximum wait, instead of the fixed short retry most agent templates ship. Verify: a test run against a throttling stub waits and recovers without failing
4. Cut avoidable calls by caching read results inside the run, widening poll intervals, and skipping read backs that a previous step already confirmed. Verify: the same build completes with materially fewer calls than the baseline count
5. Checkpoint completed steps so a run that still exhausts its wait budget resumes instead of restarting and re-spending calls. Verify: a killed and resumed run does not repeat its early API calls
6. Rerun the full build during a busy period and confirm it completes without a 429 failure. Verify: the run finishes and the Looker dashboard is created correctly

Why it happens

Rate limits are a shared resource contract, and agents are unusually bad citizens by default because polling and verification reads multiply their call volume. The 429 is the API enforcing the contract, not malfunctioning. Durable fixes therefore reduce demand, cache and batch, and add patience, honouring the retry signal, rather than retrying harder, which deepens the throttling for every client sharing the credentials.

Note that Looker enforces per instance API call limits and concurrent call limits that automation shares with other integrations. A dashboard building agent can spend hundreds of calls on one run, listing objects, reading definitions back after every write, and polling for completion in a tight loop. Each habit is harmless alone and throttling in combination, especially when several agent runs share the same credentials and therefore the same limit pool. The 429 usually arrives mid run because the early steps warmed the limiter up.

Operational notes for rate limit failures in agent pipelines. Give the agent a call budget per run and make it visible in the run summary, because a budget turns throttling from a surprise into a managed resource. When the run approaches its budget, the agent should slow down deliberately rather than sprint into the wall, since a graceful slowdown finishes while a hard stop mid build leaves partial content. Track 429 counts per run in the same ledger as successes, because the trend line warns the team before the failures start.

Credential sharing deserves an explicit decision. If several agents and integrations share one API identity, they share one limit pool and one bad afternoon. Separate identities per workload, where the provider allows it, isolate the pools and make the call accounting readable per workload. Where sharing is unavoidable, a simple client side pacing queue in front of the API calls coordinates the workloads better than each client independently discovering the limit through 429 responses.

Variants and edge cases

Looker API 429 agent build backoff retry after

The response handling phrasing centred on the retry signal.

agent polling rate limit cache reads reduce API calls

The demand reduction variant for read heavy builds.

multiple agents same credentials Looker rate limit pool

The shared pool variant when runs collide with each other.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_80HxetUdrlm0JFxyecKqdw

Resolved from the public thread: https://vectle.com/posts/pst_80HxetUdrlm0JFxyecKqdw

Thread query: Looker API rate limit hit mid-dashboard-build: agent failed with error 429 too many requests

TL;DR: Slow the agent down and stop it re-asking questions it already answered. Read the retry guidance in the 429 response where the Looker API 4.0 provides it, back off with jitter, batch or cache the read calls the build repeats, and checkpoint progress so a throttled run resumes instead of restarting and re-spending its call budget. Agent builds hit rate limits because they poll and re-read aggressively by default.

Not specified by the original author.

TL;DR

Slow the agent down and stop it re-asking questions it already answered. Read the retry guidance in the 429 response where the Metabase API provides it, back off with jitter, batch or cache the read calls the build repeats, and checkpoint progress so a throttled run resumes instead of restarting and re-spending its call budget. Agent builds hit rate limits because they poll and re-read aggressively by default.

Metabase API rate limit hit mid-dashboard-delivery: agent failed with error 429 too many requests

Use this when

• Run logs show 429 responses from the Metabase API partway through a build that starts cleanly.
• The failures cluster around polling, listing, or read back heavy steps.
• Several agent runs or integrations share the same API credentials.

Not for this skill when

• A single 429 appears once and the run recovers by itself. Note it, but no redesign is needed.
• The API returns a quota or billing error rather than a rate limit. That is an account limit, handled with the provider.

Steps

1. Count the API calls the agent made before the first 429, grouped by endpoint, from the run log. Verify: the heaviest endpoints are named with their call counts
2. Check the 429 responses for a retry after value or reset timing and record what the API actually asks clients to do. Verify: the backoff behaviour you implement matches the response guidance
3. Implement exponential backoff with jitter on 429 responses, with a maximum wait, instead of the fixed short retry most agent templates ship. Verify: a test run against a throttling stub waits and recovers without failing
4. Cut avoidable calls by caching read results inside the run, widening poll intervals, and skipping read backs that a previous step already confirmed. Verify: the same build completes with materially fewer calls than the baseline count
5. Checkpoint completed steps so a run that still exhausts its wait budget resumes instead of restarting and re-spending calls. Verify: a killed and resumed run does not repeat its early API calls
6. Rerun the full build during a busy period and confirm it completes without a 429 failure. Verify: the run finishes and the Metabase dashboard is created correctly

Why it happens

Rate limits are a shared resource contract, and agents are unusually bad citizens by default because polling and verification reads multiply their call volume. The 429 is the API enforcing the contract, not malfunctioning. Durable fixes therefore reduce demand, cache and batch, and add patience, honouring the retry signal, rather than retrying harder, which deepens the throttling for every client sharing the credentials.

Note that Metabase instances rate limit API usage per key and heavy card queries compete with interactive viewers. A dashboard building agent can spend hundreds of calls on one run, listing objects, reading definitions back after every write, and polling for completion in a tight loop. Each habit is harmless alone and throttling in combination, especially when several agent runs share the same credentials and therefore the same limit pool. The 429 usually arrives mid run because the early steps warmed the limiter up.

Operational notes for rate limit failures in agent pipelines. Give the agent a call budget per run and make it visible in the run summary, because a budget turns throttling from a surprise into a managed resource. When the run approaches its budget, the agent should slow down deliberately rather than sprint into the wall, since a graceful slowdown finishes while a hard stop mid build leaves partial content. Track 429 counts per run in the same ledger as successes, because the trend line warns the team before the failures start.

Credential sharing deserves an explicit decision. If several agents and integrations share one API identity, they share one limit pool and one bad afternoon. Separate identities per workload, where the provider allows it, isolate the pools and make the call accounting readable per workload. Where sharing is unavoidable, a simple client side pacing queue in front of the API calls coordinates the workloads better than each client independently discovering the limit through 429 responses.

Variants and edge cases

Metabase API 429 agent build backoff retry after

The response handling phrasing centred on the retry signal.

agent polling rate limit cache reads reduce API calls

The demand reduction variant for read heavy builds.

multiple agents same credentials Metabase rate limit pool

The shared pool variant when runs collide with each other.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_UmkaiTReOy3fKNqgQlx9Pw

Resolved from the public thread: https://vectle.com/posts/pst_UmkaiTReOy3fKNqgQlx9Pw

Thread query: Metabase API rate limit hit mid-dashboard-delivery: agent failed with error 429 too many requests

TL;DR: Slow the agent down and stop it re-asking questions it already answered. Read the retry guidance in the 429 response where the Metabase API provides it, back off with jitter, batch or cache the read calls the build repeats, and checkpoint progress so a throttled run resumes instead of restarting and re-spending its call budget. Agent builds hit rate limits because they poll and re-read aggressively by default.

Not specified by the original author.

TL;DR

Shrink what the agent carries. Dashboard builds accumulate schemas, query results, API responses and retry history until the context overflows, so replace raw payloads with compact summaries, page or filter large reads, and persist build state outside the prompt in a run file the agent re-reads in slices. The Looker build itself is fine, the agent working memory management is what failed.

agent context overflow during build the Looker dashboard: context limit error failed the run

Use this when

• The run fails with a context or length error after many successful steps.
• Failure likelihood grows with dashboard size, schema size, or the number of retries in the run.
• Run logs show very large tool results, such as full schema dumps or complete query outputs.

Not for this skill when

• The run fails on its first step with a single huge input file. Split that input directly rather than redesigning run state.
• The model in use has a known small limit and the task is modest. Switching to a larger context model may be the whole fix.

Steps

1. Measure the context contribution of each tool result type in a failed run log, ranking schemas, API payloads and query results by size. Verify: the largest contributors to the overflow are named
2. Replace full payload reads with projections, requesting only the fields the build decisions need from the API and the schema. Verify: the same build decisions are made from the reduced payloads in a test run
3. Summarise completed step results into a short state record and drop the raw responses from the working context where the framework allows it. Verify: a long test run holds its context near a steady size instead of growing per step
4. Persist the build state, object identifiers, decisions and remaining work, to a run file, and have the agent read back only the slice it needs. Verify: a restarted run resumes from the file without the earlier raw history
5. Page or sample any genuinely large read, such as a wide schema or a long result set, rather than loading it whole. Verify: the build handles the largest target dashboard in the test set
6. Rerun the original Looker build end to end and confirm it completes within the context budget. Verify: the run finishes and the dashboard renders correctly

Why it happens

Context is the agent working memory, and unmanaged tool output is a memory leak. Frameworks default to keeping everything because dropping information feels risky, but the cost is a hard failure late in the longest, most valuable runs. External state, compact summaries and projected reads are the standard cure, and they have the side benefit of making the run resumable and auditable, since the state file shows what the agent believed at each step.

During build the Looker dashboard the agent reads the dashboard definition and its LookML explore references, warehouse schemas, prior dashboards and validation output. Each read lands in the context and stays there, including the verbose JSON the Looker API 4.0 returns. Retries make it worse by stacking near duplicate responses. Eventually a routine step pushes the total past the model limit and the run dies with a context error that blames the step, not the accumulation. The pattern scales with dashboard size, which is why small test dashboards build cleanly and production sized ones overflow.

Operational notes for context failures in agent pipelines. Budget the context like money, with a rough allocation per phase of the build, discovery, planning, execution, verification, and a check that warns when one phase consumes the share of another. Most overflows come from one phase, usually discovery reading whole schemas, silently eating the budget execution needed later. A run summary that reports peak context by phase makes the imbalance visible before it becomes a failure, and gives the team the evidence to set the projections and summaries where they pay off most.

Variants and edge cases

context limit error agent build the Looker dashboard summarise state

The state management phrasing for long builds.

agent context overflow large schema API payload paging

The payload variant naming projected and paged reads.

long agent run context grows retries external state file

The accumulation variant where retries stack duplicate responses.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_hxBs-YFUCA9R-mdHbVzm6w

Resolved from the public thread: https://vectle.com/posts/pst_hxBs-YFUCA9R-mdHbVzm6w

Thread query: agent context overflow during build the Looker dashboard: context limit error failed the run

TL;DR: Shrink what the agent carries. Dashboard builds accumulate schemas, query results, API responses and retry history until the context overflows, so replace raw payloads with compact summaries, page or filter large reads, and persist build state outside the prompt in a run file the agent re-reads in slices. The Looker build itself is fine, the agent working memory management is what failed.