require 'arvados/keep'
+require 'sweep_trashed_collections'
class Collection < ArvadosModel
+ extend CurrentApiClient
extend DbCurrentTime
include HasUuid
include KindAndEtag
serialize :properties, Hash
+ before_validation :set_validation_timestamp
before_validation :default_empty_manifest
before_validation :check_encoding
before_validation :check_manifest_validity
before_validation :check_signatures
before_validation :strip_signatures_and_update_replication_confirmed
+ before_validation :ensure_trash_at_not_in_past
+ before_validation :sync_trash_state
+ before_validation :default_trash_interval
validate :ensure_pdh_matches_manifest_text
+ validate :validate_trash_and_delete_timing
before_save :set_file_names
- before_save :expires_at_not_in_past
- # Query only undeleted collections by default.
- default_scope where("expires_at IS NULL or expires_at > statement_timestamp()")
+ # Query only untrashed collections by default.
+ default_scope { where("is_trashed = false") }
api_accessible :user, extend: :common do |t|
t.add :name
t.add :properties
t.add :portable_data_hash
t.add :signed_manifest_text, as: :manifest_text
+ t.add :manifest_text, as: :unsigned_manifest_text
t.add :replication_desired
t.add :replication_confirmed
t.add :replication_confirmed_at
- t.add :expires_at
+ t.add :delete_at
+ t.add :trash_at
+ t.add :is_trashed
+ end
+
+ after_initialize do
+ @signatures_checked = false
+ @computed_pdh_for_manifest_text = false
end
def self.attributes_required_columns
# API response, and never let clients select the
# manifest_text column.
#
- # We need expires_at to determine the correct
- # timestamp in signed_manifest_text.
- 'manifest_text' => ['manifest_text', 'expires_at'],
+ # We need trash_at and is_trashed to determine the
+ # correct timestamp in signed_manifest_text.
+ 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
+ 'unsigned_manifest_text' => ['manifest_text'],
)
end
# subsequent passes without checking any signatures. This is
# important because the signatures have probably been stripped off
# by the time we get to a second validation pass!
- return true if @signatures_checked and @signatures_checked == computed_pdh
+ if @signatures_checked && @signatures_checked == computed_pdh
+ return true
+ end
if self.manifest_text_changed?
# Check permissions on the collection manifest.
api_token = current_api_client_authorization.andand.api_token
signing_opts = {
api_token: api_token,
- now: db_current_time.to_i,
+ now: @validation_timestamp.to_i,
}
self.manifest_text.each_line do |entry|
entry.split.each do |tok|
false
elsif portable_data_hash[0..31] != computed_pdh[0..31]
errors.add(:portable_data_hash,
- "does not match computed hash #{computed_pdh}")
+ "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
false
else
# Ignore the client-provided size part: always store
utf8 = manifest_text
utf8.force_encoding Encoding::UTF_8
if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
- manifest_text = utf8
+ self.manifest_text = utf8
return true
end
rescue
end
def signed_manifest_text
- if has_attribute? :manifest_text
+ if !has_attribute? :manifest_text
+ return nil
+ elsif is_trashed
+ return manifest_text
+ else
token = current_api_client_authorization.andand.api_token
exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
- expires_at].compact.map(&:to_i).min
- @signed_manifest_text = self.class.sign_manifest manifest_text, token, exp
+ trash_at].compact.map(&:to_i).min
+ self.class.sign_manifest manifest_text, token, exp
end
end
hash_part = nil
size_part = nil
uuid.split('+').each do |token|
- if token.match /^[0-9a-f]{32,}$/
+ if token.match(/^[0-9a-f]{32,}$/)
raise "uuid #{uuid} has multiple hash parts" if hash_part
hash_part = token
- elsif token.match /^\d+$/
+ elsif token.match(/^\d+$/)
raise "uuid #{uuid} has multiple size parts" if size_part
size_part = token
end
[hash_part, size_part].compact.join '+'
end
- # Return array of Collection objects
- def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
+ def self.get_compatible_images(readers, pattern, collections)
+ if collections.empty?
+ return []
+ end
+
+ migrations = Hash[
+ Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
+ collections.map(&:portable_data_hash),
+ 'docker_image_migration',
+ system_user_uuid).
+ order('links.created_at asc').
+ map { |l|
+ [l.tail_uuid, l.head_uuid]
+ }]
+
+ migrated_collections = Hash[
+ Collection.readable_by(*readers).
+ where('portable_data_hash in (?)', migrations.values).
+ map { |c|
+ [c.portable_data_hash, c]
+ }]
+
+ collections.map { |c|
+ # Check if the listed image is compatible first, if not, then try the
+ # migration link.
+ manifest = Keep::Manifest.new(c.manifest_text)
+ if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
+ c
+ elsif m = migrated_collections[migrations[c.portable_data_hash]]
+ manifest = Keep::Manifest.new(m.manifest_text)
+ if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
+ m
+ end
+ end
+ }.compact
+ end
+
+ # Resolve a Docker repo+tag, hash, or collection PDH to an array of
+ # Collection objects, sorted by timestamp starting with the most recent
+ # match.
+ #
+ # If filter_compatible_format is true (the default), only return image
+ # collections which are support by the installation as indicated by
+ # Rails.configuration.docker_image_formats. Will follow
+ # 'docker_image_migration' links if search_term resolves to an incompatible
+ # image, but an equivalent compatible image is available.
+ def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
readers ||= [Thread.current[:user]]
base_search = Link.
readable_by(*readers).
joins("JOIN collections ON links.head_uuid = collections.uuid").
order("links.created_at DESC")
+ if (Rails.configuration.docker_image_formats.include? 'v1' and
+ Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
+ pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
+ elsif Rails.configuration.docker_image_formats.include? 'v2'
+ pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
+ elsif Rails.configuration.docker_image_formats.include? 'v1'
+ pattern = /^[0-9A-Fa-f]{64}\.tar$/
+ else
+ raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
+ end
+
# If the search term is a Collection locator that contains one file
# that looks like a Docker image, return it.
if loc = Keep::Locator.parse(search_term)
loc.strip_hints!
- coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
- if coll_match
- # Check if the Collection contains exactly one file whose name
- # looks like a saved Docker image.
- manifest = Keep::Manifest.new(coll_match.manifest_text)
- if manifest.exact_file_count?(1) and
- (manifest.files[0][1] =~ /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/)
- return [coll_match]
- end
- end
+ coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
+ return get_compatible_images(readers, pattern, coll_match)
end
if search_tag.nil? and (n = search_term.index(":"))
# so that anything with an image timestamp is considered more recent than
# anything without; then we use the link's created_at as a tiebreaker.
uuid_timestamps = {}
- matches.all.map do |link|
+ matches.each do |link|
uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
-link.created_at.to_i]
+ end
+
+ sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
+ uuid_timestamps[c.uuid]
+ }
+ compatible = get_compatible_images(readers, pattern, sorted)
+ if sorted.length > 0 and compatible.empty?
+ raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
end
- Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
+ compatible
end
def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
super - ["manifest_text"]
end
+ def self.where *args
+ SweepTrashedCollections.sweep_if_stale
+ super
+ end
+
protected
def portable_manifest_text
self.class.munge_manifest_locators(manifest_text) do |match|
super
end
- # If expires_at is being changed to a time in the past, change it to
+ # Use a single timestamp for all validations, even though each
+ # validation runs at a different time.
+ def set_validation_timestamp
+ @validation_timestamp = db_current_time
+ end
+
+ # If trash_at is being changed to a time in the past, change it to
# now. This allows clients to say "expires {client-current-time}"
# without failing due to clock skew, while avoiding odd log entries
# like "expiry date changed to {1 year ago}".
- def expires_at_not_in_past
- if expires_at_changed? and expires_at
- self.expires_at = [db_current_time, expires_at].max
+ def ensure_trash_at_not_in_past
+ if trash_at_changed? && trash_at
+ self.trash_at = [@validation_timestamp, trash_at].max
+ end
+ end
+
+ # Caller can move into/out of trash by setting/clearing is_trashed
+ # -- however, if the caller also changes trash_at, then any changes
+ # to is_trashed are ignored.
+ def sync_trash_state
+ if is_trashed_changed? && !trash_at_changed?
+ if is_trashed
+ self.trash_at = @validation_timestamp
+ else
+ self.trash_at = nil
+ self.delete_at = nil
+ end
end
+ self.is_trashed = trash_at && trash_at <= @validation_timestamp || false
+ true
+ end
+
+ def default_trash_interval
+ if trash_at_changed? && !delete_at_changed?
+ # If trash_at is updated without touching delete_at,
+ # automatically update delete_at to a sensible value.
+ if trash_at.nil?
+ self.delete_at = nil
+ else
+ self.delete_at = trash_at + Rails.configuration.default_trash_lifetime.seconds
+ end
+ elsif !trash_at || !delete_at || trash_at > delete_at
+ # Not trash, or bogus arguments? Just validate in
+ # validate_trash_and_delete_timing.
+ elsif delete_at_changed? && delete_at >= trash_at
+ # Fix delete_at if needed, so it's not earlier than the expiry
+ # time on any permission tokens that might have been given out.
+
+ # In any case there are no signatures expiring after now+TTL.
+ # Also, if the existing trash_at time has already passed, we
+ # know we haven't given out any signatures since then.
+ earliest_delete = [
+ @validation_timestamp,
+ trash_at_was,
+ ].compact.min + Rails.configuration.blob_signature_ttl.seconds
+
+ # The previous value of delete_at is also an upper bound on the
+ # longest-lived permission token. For example, if TTL=14,
+ # trash_at_was=now-7, delete_at_was=now+7, then it is safe to
+ # set trash_at=now+6, delete_at=now+8.
+ earliest_delete = [earliest_delete, delete_at_was].compact.min
+
+ # If delete_at is too soon, use the earliest possible time.
+ if delete_at < earliest_delete
+ self.delete_at = earliest_delete
+ end
+ end
+ end
+
+ def validate_trash_and_delete_timing
+ if trash_at.nil? != delete_at.nil?
+ errors.add :delete_at, "must be set if trash_at is set, and must be nil otherwise"
+ elsif delete_at && delete_at < trash_at
+ errors.add :delete_at, "must not be earlier than trash_at"
+ end
+ true
end
end