OpenAgents

Commit

Fix boot convergence for rolling images

ac40f633af5c · AtlantisPleb · · parent 169f8c3cab77 

Fix boot convergence for rolling images

Deploy story

What this commit did to the running system — joined from the forge receipt chain, the part a commit page elsewhere cannot show.

Not deployed through the forge lane

No push, promotion, build, or deploy receipt references this commit (receipts are scanned over a bounded recent window). Changes shipped by full node replacement carry their proof in the release gate receipt instead.

Changed files

  • modified lib/openagents/forge/boot_converge.ex
  • modified test/openagents/forge/boot_converge_test.exs

Diff

2 files changed, +99 -21

lib/openagents/forge/boot_converge.ex modified +38 -21

@@ -164,7 +164,7 @@ defmodule OpenAgents.Forge.BootConverge do

164 164 
        true
165 165
166 166 
      {%{sha: sha, details: details}, %{"sha" => sha, "artifact_digest" => nil}} ->
167 
        (details || %{})["artifact_digest"] == nil and OpenAgents.BuildInfo.revision() == sha
167 
        image_state_matches?(sha, details || %{})
168 168
169 169 
      {%{sha: sha, details: details},
170 170 
       %{

@@ -193,31 +193,48 @@ defmodule OpenAgents.Forge.BootConverge do

193 193 
  end
194 194
195 195 
  defp converge_target(repo, target, sha, details, attempts) do
196 
    with {:ok, identity} <- target_identity(target, details),
197 
         {:ok, bytes, cache_state} <- artifact_bytes(repo, identity),
198 
         {:ok, response} <- DeploymentNode.install_artifact(install_request(identity, bytes)),
199 
         :ok <- retain_artifacts(repo, target.id, identity.artifact_digest) do
200 
      %{
201 
        "schema" => "openagents.forge.boot-convergence.v2",
202 
        "state" => "converged",
203 
        "ready" => true,
204 
        "reason" => cache_state,
205 
        "sha" => sha,
206 
        "artifact_digest" => identity.artifact_digest,
207 
        "manifest_digest" => identity.manifest_digest,
208 
        "modules" => response["modules"] || identity.modules,
209 
        "attempts" => attempts,
210 
        "retry_in_ms" => nil
211 
      }
196 
    if image_target_matches?(sha, details) do
197 
      image_ready("image_matches_live", attempts, sha)
212 198 
    else
213 
      {:image_matches, ^sha} ->
214 
        image_ready("image_matches_live", attempts, sha)
199 
      with {:ok, identity} <- target_identity(target, details),
200 
           {:ok, bytes, cache_state} <- artifact_bytes(repo, identity),
201 
           {:ok, response} <- DeploymentNode.install_artifact(install_request(identity, bytes)),
202 
           :ok <- retain_artifacts(repo, target.id, identity.artifact_digest) do
203 
        %{
204 
          "schema" => "openagents.forge.boot-convergence.v2",
205 
          "state" => "converged",
206 
          "ready" => true,
207 
          "reason" => cache_state,
208 
          "sha" => sha,
209 
          "artifact_digest" => identity.artifact_digest,
210 
          "manifest_digest" => identity.manifest_digest,
211 
          "modules" => response["modules"] || identity.modules,
212 
          "attempts" => attempts,
213 
          "retry_in_ms" => nil
214 
        }
215 
      else
216 
        {:image_matches, ^sha} ->
217 
          image_ready("image_matches_live", attempts, sha)
215 218
216 
      {:error, reason} ->
217 
        degraded(OpenAgents.OperationalLog.code(reason), attempts, reason, sha)
219 
        {:error, reason} ->
220 
          degraded(OpenAgents.OperationalLog.code(reason), attempts, reason, sha)
221 
      end
218 222 
    end
219 223 
  end
220 224
225 
  defp image_state_matches?(sha, details) do
226 
    (details["artifact_digest"] == nil and OpenAgents.BuildInfo.revision() == sha) or
227 
      image_target_matches?(sha, details)
228 
  end
229
230 
  defp image_target_matches?(sha, details) do
231 
    runtime_digest = OpenAgents.BuildInfo.image_digest()
232
233 
    is_binary(runtime_digest) and
234 
      OpenAgents.BuildInfo.revision() == sha and
235 
      details["image_digest"] == runtime_digest
236 
  end
237
221 238 
  defp target_identity(target, details) do
222 239 
    case details do
223 240 
      %{
test/openagents/forge/boot_converge_test.exs modified +61

@@ -348,6 +348,67 @@ defmodule OpenAgents.Forge.BootConvergeTest do

348 348 
    assert BootConverge.ready?(@repo)
349 349 
  end
350 350
351 
  test "image-matching rolling target remains ready despite a non-direct artifact" do
352 
    runtime_sha = OpenAgents.BuildInfo.revision()
353 
    runtime_digest = "sha256:" <> String.duplicate("f", 64)
354 
    previous_digest = Application.get_env(:openagents, :image_digest)
355 
    Application.put_env(:openagents, :image_digest, runtime_digest)
356
357 
    on_exit(fn -> restore_env(:image_digest, previous_digest) end)
358
359 
    {module, binary} = scratch_beam(OpenAgents.NotAllowed.BootConvergeRollingImage)
360 
    artifact = artifact(module, binary)
361
362 
    target =
363 
      insert_target!(
364 
        "live",
365 
        Map.put(artifact.details, "image_digest", runtime_digest)
366 
      )
367
368 
    target
369 
    |> Ecto.Changeset.change(%{sha: runtime_sha})
370 
    |> Repo.update!()
371
372 
    previous_enabled = Application.get_env(:openagents, :forge_boot_converge_enabled)
373 
    Application.put_env(:openagents, :forge_boot_converge_enabled, true)
374
375 
    on_exit(fn -> restore_env(:forge_boot_converge_enabled, previous_enabled) end)
376
377 
    assert %{
378 
             "state" => "image",
379 
             "ready" => true,
380 
             "reason" => "image_matches_live",
381 
             "sha" => ^runtime_sha
382 
           } = BootConverge.converge(@repo)
383
384 
    assert BootConverge.ready?(@repo)
385 
    refute Code.ensure_loaded?(module)
386 
  end
387
388 
  test "rolling target stays degraded when its image digest does not match the runtime" do
389 
    runtime_sha = OpenAgents.BuildInfo.revision()
390 
    previous_digest = Application.get_env(:openagents, :image_digest)
391 
    Application.put_env(:openagents, :image_digest, "sha256:" <> String.duplicate("f", 64))
392
393 
    on_exit(fn -> restore_env(:image_digest, previous_digest) end)
394
395 
    {module, binary} = scratch_beam(OpenAgents.NotAllowed.BootConvergeWrongImage)
396 
    artifact = artifact(module, binary)
397 
    File.write!(Path.join(Repos.data_dir(), artifact.details["artifact"]), artifact.built.bytes)
398
399 
    target =
400 
      insert_target!(
401 
        "live",
402 
        Map.put(artifact.details, "image_digest", "sha256:" <> String.duplicate("e", 64))
403 
      )
404
405 
    target
406 
    |> Ecto.Changeset.change(%{sha: runtime_sha})
407 
    |> Repo.update!()
408
409 
    assert %{"state" => "degraded", "ready" => false} = BootConverge.converge(@repo)
410 
  end
411
351 412 
  test "an unreadable cache entry degrades with a bounded reason" do
352 413 
    {module, binary} = scratch_beam(OpenAgents.Scratch.BootConvergeUnreadable)
353 414 
    artifact = artifact(module, binary)

This page updates live while a promote is in flight · changelog