bridgy-fed/atproto_hub.yaml

# https://cloud.google.com/appengine/docs/flexible/reference/app-yaml?tab=python
#
# gcloud -q app deploy --project bridgy-federated hub.yaml

# application: bridgy-federated

service: atproto-hub
env: flex
runtime: python
runtime_config:
    operating_system: ubuntu22
    runtime_version: "3.12"

resources:
  cpu: 1
  # steady state memory usage is only .5-2G, but spikes when we're working
  # through a backlog of commits
  # https://github.com/snarfed/bridgy-fed/issues/1266#issuecomment-2321863960
  memory_gb: 6.0

network:
  # run in default VPC for this project
  # https://cloud.google.com/appengine/docs/flexible/using-shared-vpc
  # https://cloud.google.com/appengine/docs/flexible/reference/app-yaml#network_settings
  # https://cloud.google.com/vpc/docs/serverless-vpc-access#supported_services
  name: projects/bridgy-federated/global/networks/default
  # can't be internal because Bluesky relay(s) need to be able to connect to us
  # over websocket for subscribeRepos
  instance_ip_mode: external

env_variables:
  PDS_HOST: atproto.brid.gy
  PLC_HOST: plc.directory
  APPVIEW_HOST: api.bsky.app
  BGS_HOST: bsky.network
  MOD_SERVICE_HOST: mod.bsky.app
  MOD_SERVICE_DID: did:plc:ar7c4by46qjdydhdevvrndac
  # ...or test against labeler.dholms.xyz / did:plc:vzxheqfwpbi3lxbgdh22js66
  MEMCACHE_HOST: '10.126.144.3'
  ROLLBACK_WINDOW: 50000
  SUBSCRIBE_REPOS_BATCH_DELAY: 10

  # https://github.com/googleapis/python-ndb/blob/c55ec62b5153787404488b046c4bf6ffa02fee64/google/cloud/ndb/utils.py#L78-L81
  # NDB_DEBUG: true

# need only one instance so that new commits can be delivered to subscribeRepos
# subscribers in memory
manual_scaling:
  instances: 1

liveness_check:
  # default 300; does lowering this speed up deploy? seems like maybe, a bit?
  # https://cloud.google.com/appengine/docs/flexible/reference/app-yaml?tab=python#liveness_checks
  initial_delay_sec: 30

# https://cloud.google.com/appengine/docs/flexible/python/runtime#application_startup
# https://docs.gunicorn.org/en/latest/settings.html#timeout
# TODO: try asyncio w/eventlet workers
entrypoint: gunicorn --workers 1 --threads 25 -b :$PORT atproto_hub:app
start on new hub service for ATProto and eventually Nostr websockets 2023-09-05 19:06:07 +00:00			`# https://cloud.google.com/appengine/docs/flexible/reference/app-yaml?tab=python`
			`#`
			`# gcloud -q app deploy --project bridgy-federated hub.yaml`

			`# application: bridgy-federated`

merge hub and atproto_firehose services into a single service, atproto_hub atproto_firehose is averaging 20-40% CPU right now, hub 5-10%, and both are 500-600M memory steady state, so they can easily fit together. 2024-06-13 00:04:59 +00:00			`service: atproto-hub`
start on new hub service for ATProto and eventually Nostr websockets 2023-09-05 19:06:07 +00:00			`env: flex`
			`runtime: python`
			`runtime_config:`
			`operating_system: ubuntu22`
upgrade App Engine app runtime to python 3.12 https://cloud.google.com/appengine/docs/standard/reference/app-yaml?tab=python#runtime https://cloud.google.com/appengine/docs/flexible/python/runtime#recommended_gunicorn_configuration 2024-09-10 04:58:52 +00:00			`runtime_version: "3.12"`
hub.yaml: minor tweaks to get service running 2023-09-11 04:54:28 +00:00
infra: add new router flex service, switch tasks to run on it receive, send, webmention, and poll-feed 2024-05-24 03:40:54 +00:00			`resources:`
atproto-hub: drop back down to 1 core 2025-01-02 01:02:58 +00:00			`cpu: 1`
bump atproto-hub up to 2 cores for snarfed/arroba#39 2024-12-19 20:50:13 +00:00			`# steady state memory usage is only .5-2G, but spikes when we're working`
hub: bump memory up to 4G, subscribeRepos rollback window to 100k seqs for #1266 2024-11-17 18:02:31 +00:00			`# through a backlog of commits`
			`# https://github.com/snarfed/bridgy-fed/issues/1266#issuecomment-2321863960`
router back up to 10 cores, atproto-hub to 6G memory 2024-11-21 23:54:18 +00:00			`memory_gb: 6.0`
bump up hub to 2G memory, explicitly GC during polls 2024-05-04 20:28:12 +00:00
hub: add dispatch.yaml for URL-based routing, update network ingress 2023-09-29 21:28:36 +00:00			`network:`
run router and atproto-hub in default VPC https://cloud.google.com/appengine/docs/flexible/using-shared-vpc https://cloud.google.com/appengine/docs/flexible/reference/app-yaml#network_settings https://cloud.google.com/vpc/docs/serverless-vpc-access#supported_services for #1149 2024-12-20 20:47:07 +00:00			`# run in default VPC for this project`
			`# https://cloud.google.com/appengine/docs/flexible/using-shared-vpc`
			`# https://cloud.google.com/appengine/docs/flexible/reference/app-yaml#network_settings`
			`# https://cloud.google.com/vpc/docs/serverless-vpc-access#supported_services`
			`name: projects/bridgy-federated/global/networks/default`
			`# can't be internal because Bluesky relay(s) need to be able to connect to us`
			`# over websocket for subscribeRepos`
hub: add dispatch.yaml for URL-based routing, update network ingress 2023-09-29 21:28:36 +00:00			`instance_ip_mode: external`
start on new hub service for ATProto and eventually Nostr websockets 2023-09-05 19:06:07 +00:00
use prod Bluesky env vars in main app, sandbox in hub 2023-09-29 18:41:19 +00:00			`env_variables:`
atproto: switch PDS hostname to atproto.brid.gy matches the hostname that's currently whitelisted in the federation beta test 2024-03-11 22:30:28 +00:00			`PDS_HOST: atproto.brid.gy`
switch hub from Bluesky sandbox to prod! 2024-02-28 18:57:46 +00:00			`PLC_HOST: plc.directory`
			`APPVIEW_HOST: api.bsky.app`
			`BGS_HOST: bsky.network`
ATroto.send: for flag activities, send createReport to mod service for snarfed/granary#596. background: * https://docs.bsky.app/docs/advanced-guides/moderation#reporting * https://github.com/bluesky-social/atproto/discussions/2350#discussioncomment-8847613 * https://github.com/snarfed/granary/issues/596 2024-04-26 20:59:04 +00:00			`MOD_SERVICE_HOST: mod.bsky.app`
			`MOD_SERVICE_DID: did:plc:ar7c4by46qjdydhdevvrndac`
			`# ...or test against labeler.dholms.xyz / did:plc:vzxheqfwpbi3lxbgdh22js66`
cache models.get_original_user/object_key queries in memcache adds new generic common.memcache_memoize decorator for caching any function's output in memcache. for #1149 2024-12-12 04:17:36 +00:00			`MEMCACHE_HOST: '10.126.144.3'`
atproto-hub: drop subscribeRepos rollback window back down to 50k seqs roughly 12-18h. for snarfed/arroba#30, snarfed/arroba#39 2024-12-19 19:36:18 +00:00			`ROLLBACK_WINDOW: 50000`
atproto-hub: set 10s subscribeRepos batch delay see https://github.com/snarfed/arroba/issues/36 2024-11-05 18:45:23 +00:00			`SUBSCRIBE_REPOS_BATCH_DELAY: 10`
set up, but don't enable, ndb debug logging for #1149 2024-12-19 00:00:20 +00:00
temporarily enable debug level ndb logging for #1149 https://github.com/googleapis/python-ndb/blob/c55ec62b5153787404488b046c4bf6ffa02fee64/google/cloud/ndb/utils.py#L78-L81 2024-12-18 23:15:46 +00:00			`# https://github.com/googleapis/python-ndb/blob/c55ec62b5153787404488b046c4bf6ffa02fee64/google/cloud/ndb/utils.py#L78-L81`
turn google.cloud logging propagate on; fixes ndb logs! google.cloud logging propagate is why I haven't been able to get ndb logs to show up! phew, finally found it. https://github.com/googleapis/python-logging/blob/8c2c5a798aa444ffa71d82a12428fbcd50395a7e/google/cloud/logging_v2/handlers/handlers.py#L302-L314 2025-01-13 20:43:02 +00:00			`# NDB_DEBUG: true`
use prod Bluesky env vars in main app, sandbox in hub 2023-09-29 18:41:19 +00:00
start on new hub service for ATProto and eventually Nostr websockets 2023-09-05 19:06:07 +00:00			`# need only one instance so that new commits can be delivered to subscribeRepos`
			`# subscribers in memory`
			`manual_scaling:`
			`instances: 1`

add /admin/atproto for rendering current subscribeRepos clients 2024-07-14 15:04:47 +00:00			`liveness_check:`
app configs: leaning into cargo-culting liveness_check.initial_delay_sec https://cloud.google.com/appengine/docs/flexible/reference/app-yaml?tab=python#liveness_checks 2024-07-15 15:46:35 +00:00			`# default 300; does lowering this speed up deploy? seems like maybe, a bit?`
add /admin/atproto for rendering current subscribeRepos clients 2024-07-14 15:04:47 +00:00			`# https://cloud.google.com/appengine/docs/flexible/reference/app-yaml?tab=python#liveness_checks`
			`initial_delay_sec: 30`

start on new hub service for ATProto and eventually Nostr websockets 2023-09-05 19:06:07 +00:00			`# https://cloud.google.com/appengine/docs/flexible/python/runtime#application_startup`
			`# https://docs.gunicorn.org/en/latest/settings.html#timeout`
			`# TODO: try asyncio w/eventlet workers`
merge hub and atproto_firehose services into a single service, atproto_hub atproto_firehose is averaging 20-40% CPU right now, hub 5-10%, and both are 500-600M memory steady state, so they can easily fit together. 2024-06-13 00:04:59 +00:00			`entrypoint: gunicorn --workers 1 --threads 25 -b :$PORT atproto_hub:app`