- set memory limit to 1.2x memory request to provide extra padding and
avoid OOM
- attempt to resize crawler pods by 1.2x when exceeding 90% of available
memory
- do a 'soft OOM' (send extra SIGTERM) to pod when reaching 100% of
requested memory, resulting in faster graceful restart, but avoiding a
system-instant OOM Kill
- Fixes#1632
---------
Co-authored-by: Tessa Walsh <tessa@bitarchivist.net>
* keep track of per pod status on crawljob:
- crashes time, and reason
- 'used' vs 'allocated' resources
- 'percent' used / allocated
* crawl log errors: log error when crawler crashes via OOM, either via redis error log
or to console
* add initial autoscaling support!
- detect if metrics server is available via K8SApi.is_pod_metrics_available()
- if available, use metrics for 'used' fields
- if no metrics, set memory used for redis only (using redis apis)
- allow overriding memory and cpu via newMemory and newCpu settings on pod status
- scale memory / cpu based on newMemory and newCpu setting
- templates: update jinja templates to allow restarting crawler and redis with new resources
- ci: enable metrics-server on k3d, microk8s and nightly k3d ci runs
* roles: cleanup unused roles, add permissions for listing metrics
* stats for running crawls:
- update in db via operator
- avoids losing stats if redis pod happens to be done
- tradeoff is more db access in operator, but less extra connections to redis + already
loading from db in backend
- size stat: ensure size of previous files is added to the stats
* crawler deployment tweaks:
- adjust cpu/mem per browser
- add --headless flag to configmap to use new headless mode by default!
* Btrixjobs Operator - Phase 1 (#679)
- add metacontroller and custom crds
- add main_op entrypoint for operator
* Btrix Operator Crawl Management (#767)
* operator backend:
- run operator api in separate container but in same pod, with WEB_CONCURRENCY=1
- operator creates statefulsets and services for CrawlJob and ProfileJob
- operator: use service hook endpoint, set port in values.yaml
* crawls working with CrawlJob
- jobs start with 'crawljob-' prefix
- update status to reflect current crawl state
- set sync time to 10 seconds by default, overridable with 'operator_resync_seconds'
- mark crawl as running, failed, complete when finished
- store finished status when crawl is complete
- support updating scale, forcing rollover, stop via patching CrawlJob
- support cancel via deletion
- requires hack to content-length for patching custom resources
- auto-delete of CrawlJob via 'ttlSecondsAfterFinished'
- also delete pvcs until autodelete supported via statefulset (k8s >1.27)
- ensure filesAdded always set correctly, keep counter in redis, add to status display
- optimization: attempt to reduce automerging, by reusing volumeClaimTemplates from existing children, as these may have additional props added
- add add_crawl_errors_to_db() for storing crawl errors from redis '<crawl>:e' key to mongodb when crawl is finished/failed/canceled
- add .status.size to display human-readable crawl size, if available (from webrecorder/browsertrix-crawler#291)
- support new page size, >0.9.0 and old page size key (changed in webrecorder/browsertrix-crawler#284)
* support for scheduled jobs!
- add main_scheduled_job entrypoint to run scheduled jobs
- add crawl_cron_job.yaml template for declaring CronJob
- CronJobs moved to default namespace
* operator manages ProfileJobs:
- jobs start with 'profilejob-'
- update expiry time by updating ProfileJob object 'expireTime' while profile is active
* refactor/cleanup:
- remove k8s package
- merge k8sman and basecrawlmanager into crawlmanager
- move templates, k8sapi, utils into root package
- delete all *_job.py files
- remove dt_now, ts_now from crawls, now in utils
- all db operations happen in crawl/crawlconfig/org files
- move shared crawl/crawlconfig/org functions that use the db to be importable directly,
including get_crawl_config, add_new_crawl, inc_crawl_stats
* role binding: more secure setup, don't allow crawler namespace any k8s permissions
- move cronjobs to be created in default namespace
- grant default namespace access to create cronjobs in default namespace
- remove role binding from crawler namespace
* additional tweaks to templates:
- templates: split crawler and redis statefulset into separate yaml file (in case need to load one or other separately)
* stats / redis optimization:
- don't update stats in mongodb on every operator sync, only when crawl is finished
- for api access, read stats directly from redis to get up-to-date stats
- move get_page_stats() to utils, add get_redis_url() to k8sapi to unify access
* Add migration for operator changes
- Update configmap for crawl configs with scale > 1 or
crawlTimeout > 0 and schedule exists to recreate CronJobs
- add option to rerun last migration, enabled via env var and by running helm with --set=rerun_last_migration=1
* subcharts: move crawljob and profilejob crds to separate subchart, as this seems best way to guarantee proper install order with + update on upgrade with helm, add built btrix-crds-0.1.0.tgz subchart
- metacontroller: use release from ghcr, add metacontroller-helm-v4.10.1.tgz subchart
* backend api fixes
- ensure changing scale of crawl also updates it in the db
- crawlconfigs: add 'currCrawlSize' and 'lastCrawlSize' to crawlconfig api
---------
Co-authored-by: D. Lee <leepro@gmail.com>
Co-authored-by: Tessa Walsh <tessa@bitarchivist.net>
- use statefulsets instead of deployments for mongo, redis, signer
- use k8s job + statefulset for running crawls
- use separate statefulset for crawl (scaled) and single-replica redis stateful set
- move crawl job update login to crawl_updater
- remove shared redis chart
package refactor:
- move to shared code to 'btrixcloud'
- move k8s to 'btrixcloud.k8s'
- move docker to 'btrixcloud.docker'
* backend: k8s:
- support crawls with multiple wacz files, don't assume crawl complete after first wacz uploaded
- if crawl is running and has wacz file, still show as running
- k8s: allow configuring node selector for main pods (eg. nodeType=main) and for crawlers (eg. nodeType=crawling)
- profiles: support uploading to alternate storage specified via 'shared_profile_storage' value is set
- misc fixes for profiles
* backend: ensure docker run_profile api matches k8s
k8s chart: don't delete pvc and pv in helm chart
* dependency: bump authsign to 0.4.0
docker: disable public redis port
* profiles: fix path, profile browser return value
* fix typo in presigned url cacheing
- add liveness probe for crawls, configurable via 'crawler_liveness_port'
- add User system:anonymous permissions
- treat jobs that have exceeded total as 'partial_complete' (experimental)
- Add default vs custom (s3) storage
- K8S: All storages correspond to secrets
- K8S: Default storages inited via helm
- K8S: Custom storage results in custom secret (per archive)
- K8S: Don't add secret per crawl config
- API for changing storage per archive
- Docker: default storage just hard-coded from env vars (only one for now)
- Validate custom storage via aiobotocore before confirming
- Data Model: remove usage from users
- Data Model: support adding multiple files per crawl for parallel crawls
- Data Model: track completions for parallel crawls
- Data Model: initial support for tags per crawl, add collection as 'coll' tag
README fixes
- replace storages with archives, which have a single storage (for now)
- crawls associated with archives
- users below to archive, with one admin user (if archive created by default)
- update crawlconfig for latest browsertrix-crawler (0.4.4)
- k8s: fix permissions for crawler role
- k8s: fix minio service (now requiring two ports)
- working apis for adding crawls, removing crawls in mongo, mapped to k8s cronjobs
- more complete crawl spec
- option to start on-demand job from cronjobs
- optional minio in separate deployment/service
