module Redwood
class Index
+ class LockError < StandardError
+ def initialize h
+ @h = h
+ end
+
+ def method_missing m; @h[m.to_s] end
+ end
+
include Singleton
attr_reader :index
+ alias ferret index
def initialize dir=BASE_DIR
@dir = dir
@sources = {}
@sources_dirty = false
wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
- sa = Ferret::Analysis::StandardAnalyzer.new Ferret::Analysis::FULL_ENGLISH_STOP_WORDS, true
+ sa = Ferret::Analysis::StandardAnalyzer.new [], true
@analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
@analyzer[:body] = sa
- @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer
+ @analyzer[:subject] = sa
+ @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
+ @lock = Lockfile.new lockfile, :retries => 0, :max_age => nil
self.class.i_am_the_instance self
end
+ def lockfile; File.join @dir, "lock" end
+
+ def lock
+ Redwood::log "locking #{lockfile}..."
+ begin
+ @lock.lock
+ rescue Lockfile::MaxTriesLockError
+ raise LockError, @lock.lockinfo_on_disk
+ end
+ end
+
+ def start_lock_update_thread
+ @lock_update_thread = Redwood::reporting_thread do
+ while true
+ sleep 30
+ @lock.touch_yourself
+ end
+ end
+ end
+
+ def stop_lock_update_thread
+ @lock_update_thread.kill if @lock_update_thread
+ @lock_update_thread = nil
+ end
+
+ def fancy_lock_error_message_for e
+ secs = Time.now - e.mtime
+ mins = secs.to_i / 60
+ time =
+ if mins == 0
+ "#{secs.to_i} seconds"
+ else
+ "#{mins} minutes"
+ end
+
+ <<EOS
+Error: the sup index is locked by another process! User '#{e.user}' on
+host '#{e.host}' is running #{e.pname} with pid #{e.pid}. The process was alive
+as of #{time} ago.
+EOS
+ end
+
+ def lock_or_die
+ begin
+ lock
+ rescue LockError => e
+ $stderr.puts fancy_lock_error_message_for(e)
+ $stderr.puts <<EOS
+
+You can wait for the process to finish, or, if it crashed and left a
+stale lock file behind, you can manually delete #{@lock.path}.
+EOS
+ exit
+ end
+ end
+
+ def unlock
+ if @lock && @lock.locked?
+ Redwood::log "unlocking #{lockfile}..."
+ @lock.unlock
+ end
+ end
+
def load
load_sources
load_index
end
def save
+ Redwood::log "saving index and sources..."
FileUtils.mkdir_p @dir unless File.exists? @dir
save_sources
save_index
end
end
- ## Update the message state on disk, by deleting and re-adding it.
- ## The message must exist in the index. docid and entry are found
- ## unless given.
- def update_message m, docid=nil, entry=nil
- unless docid && entry
- docid, entry = load_entry_for_id m.id
- raise ArgumentError, "cannot find #{m.id} in the index" unless entry
- end
-
- raise "no entry and no source info for message #{m.id}" unless m.source && m.source_info
+ ## Syncs the message to the index: deleting if it's already there,
+ ## and adding either way. Index state will be determined by m.labels.
+ ##
+ ## docid and entry can be specified if they're already known.
+ def sync_message m, docid=nil, entry=nil
+ docid, entry = load_entry_for_id m.id unless docid && entry
- raise "deleting non-corresponding entry #{docid}" unless @index[docid][:message_id] == m.id
+ raise "no source info for message #{m.id}" unless m.source && m.source_info
+ raise "trying to delete non-corresponding entry #{docid} with index message-id #{@index[docid][:message_id].inspect} and parameter message id #{m.id.inspect}" if docid && @index[docid][:message_id] != m.id
- @index.delete docid
- add_message m
- end
-
- ## for each new message form the source, yields a bunch of stuff,
- ## gets the message back from the block, and adds it or updates it.
- def add_new_messages_from source
- found = {}
- return if source.done? || source.broken?
-
- source.each do |offset, labels|
- if source.broken?
- Redwood::log "error loading messages from #{source}: #{source.broken_msg}"
- return
+ source_id =
+ if m.source.is_a? Integer
+ m.source
+ else
+ m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
end
-
- labels.each { |l| LabelManager << l }
-
- begin
- m = Message.new :source => source, :source_info => offset, :labels => labels
- if found[m.id]
- puts "skipping duplicate message #{m.id}"
- next
- else
- found[m.id] = true
- end
- if m.source_marked_read?
- m.remove_label :unread
- labels.delete :unread
- end
+ to = (m.to + m.cc + m.bcc).map { |x| x.email }.join(" ")
+ d = {
+ :message_id => m.id,
+ :source_id => source_id,
+ :source_info => m.source_info,
+ :date => m.date.to_indexable_s,
+ :body => m.content,
+ :snippet => m.snippet,
+ :label => m.labels.join(" "),
+ :from => m.from ? m.from.email : "",
+ :to => (m.to + m.cc + m.bcc).map { |x| x.email }.join(" "),
+ :subject => wrap_subj(Message.normalize_subj(m.subj)),
+ :refs => (m.refs + m.replytos).uniq.join(" "),
+ }
- docid, entry = load_entry_for_id m.id
- m = yield m, offset, labels, entry
- next unless m
- if entry
- update_message m, docid, entry
- else
- add_message m
- UpdateManager.relay :add, m
- end
- rescue MessageFormatError, SourceError => e
- Redwood::log "ignoring erroneous message at #{source}##{offset}: #{e.message}"
- end
- end
+ @index.delete docid if docid
+ @index.add_document d
+
+ docid, entry = load_entry_for_id m.id
+ ## this hasn't been triggered in a long time. TODO: decide whether it's still a problem.
+ raise "just added message #{m.id.inspect} but couldn't find it in a search" unless docid
+ true
end
def save_index fn=File.join(@dir, "ferret")
- # don't have to do anything, apparently
+ # don't have to do anything, apparently
end
def contains_id? id
def size; @index.size; end
## you should probably not call this on a block that doesn't break
- ## rather quickly because the results will probably be, as we say
- ## in scotland, frikkin' huuuge.
+ ## rather quickly because the results can be very large.
EACH_BY_DATE_NUM = 100
def each_id_by_date opts={}
return if @index.size == 0 # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
def num_results_for opts={}
return 0 if @index.size == 0 # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
+
q = build_query opts
- index.search(q).total_hits
+ index.search(q, :limit => 1).total_hits
end
## yield all messages in the thread containing 'm' by repeatedly
- ## querying the index. yields pairs of message ids and
+ ## querying the index. yields pairs of message ids and
## message-building lambdas, so that building an unwanted message
## can be skipped in the block if desired.
+ ##
+ ## only two options, :limit and :skip_killed. if :skip_killed is
+ ## true, stops loading any thread if a message with a :killed flag
+ ## is found.
SAME_SUBJECT_DATE_LIMIT = 7
def each_message_in_thread_for m, opts={}
+ #Redwood::log "Building thread for #{m.id}: #{m.subj}"
messages = {}
searched = {}
num_queries = 0
- ## temporarily disabling subject searching because it's a
- ## significant slowdown.
- ##
- ## TODO: make this configurable, i guess
- if true
+ if $config[:thread_by_subject] # do subject queries
date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
sq.add_term t
end
q.add_query sq, :must
- q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
+ q = build_query :qobj => q
+
pending = @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] }
Redwood::log "found #{pending.size} results for subject query #{q}"
else
q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
+ q = build_query :qobj => q
+
num_queries += 1
+ killed = false
@index.search_each(q, :limit => :all) do |docid, score|
break if opts[:limit] && messages.size >= opts[:limit]
+ if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
+ killed = true
+ break
+ end
mid = @index[docid][:message_id]
- unless messages.member? mid
+ unless messages.member?(mid)
+ #Redwood::log "got #{mid} as a child of #{id}"
messages[mid] ||= lambda { build_message docid }
refs = @index[docid][:refs].split(" ")
pending += refs
end
end
end
- Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}" if num_queries > 0
- messages.each { |mid, builder| yield mid, builder }
+ if killed
+ Redwood::log "thread for #{m.id} is killed, ignoring"
+ false
+ else
+ Redwood::log "ran #{num_queries} queries to build thread of #{messages.size + 1} messages for #{m.id}: #{m.subj}" if num_queries > 0
+ messages.each { |mid, builder| yield mid, builder }
+ true
+ end
end
## builds a message object from a ferret result
"from" => doc[:from],
"to" => doc[:to],
"message-id" => doc[:message_id],
- "references" => doc[:refs],
+ "references" => doc[:refs].split(/\s+/).map { |x| "<#{x}>" }.join(" "),
}
Message.new :source => source, :source_info => doc[:source_info].to_i,
def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end
- ## Adds a message to the index. The message cannot already exist in
- ## the index.
- def add_message m
- raise ArgumentError, "index already contains #{m.id}" if contains? m
-
- source_id =
- if m.source.is_a? Integer
- m.source
- else
- m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
- end
-
- to = (m.to + m.cc + m.bcc).map { |x| x.email }.join(" ")
- d = {
- :message_id => m.id,
- :source_id => source_id,
- :source_info => m.source_info,
- :date => m.date.to_indexable_s,
- :body => m.content,
- :snippet => m.snippet,
- :label => m.labels.join(" "),
- :from => m.from ? m.from.email : "",
- :to => (m.to + m.cc + m.bcc).map { |x| x.email }.join(" "),
- :subject => wrap_subj(Message.normalize_subj(m.subj)),
- :refs => (m.refs + m.replytos).uniq.join(" "),
- }
-
- @index.add_document d
-
- docid, entry = load_entry_for_id m.id
- ## this hasn't been triggered in a long time. TODO: decide whether it's still a problem.
- raise "just added message #{m.id} but couldn't find it in a search" unless docid
- true
- end
-
def drop_entry docno; @index.delete docno; end
def load_entry_for_id mid
num = h[:num] || 20
@index.search_each(q, :sort => "date DESC", :limit => :all) do |docid, score|
break if contacts.size >= num
- #Redwood::log "got message with to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
+ #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
f = @index[docid][:from]
t = @index[docid][:to]
if AccountManager.is_account_email? f
- t.split(" ").each { |e| #Redwood::log "adding #{e} because there's a message to him from account email #{f}";
- contacts[Person.for(e)] = true }
+ t.split(" ").each { |e| contacts[PersonManager.person_for(e)] = true }
else
- #Redwood::log "adding from #{f} because there's a message from him to #{t}"
- contacts[Person.for(f)] = true
+ contacts[PersonManager.person_for(f)] = true
end
end
contacts.keys.compact
end
+ def load_sources fn=Redwood::SOURCE_FN
+ source_array = (Redwood::load_yaml_obj(fn) || []).map { |o| Recoverable.new o }
+ @sources = Hash[*(source_array).map { |s| [s.id, s] }.flatten]
+ @sources_dirty = false
+ end
+
+ def has_any_from_source_with_label? source, label
+ q = Ferret::Search::BooleanQuery.new
+ q.add_query Ferret::Search::TermQuery.new("source_id", source.id.to_s), :must
+ q.add_query Ferret::Search::TermQuery.new("label", label.to_s), :must
+ index.search(q, :limit => 1).total_hits > 0
+ end
+
protected
- def parse_user_query_string str; @qparser.parse str; end
+ def parse_user_query_string str
+ str2 = str.gsub(/(to|from):(\S+)/) do
+ field, name = $1, $2
+ if(p = ContactManager.contact_for(name))
+ [field, p.email]
+ else
+ [field, name]
+ end.join(":")
+ end
+
+ Redwood::log "translated #{str} to #{str2}" unless str2 == str
+ @qparser.parse str2
+ end
+
def build_query opts
query = Ferret::Search::BooleanQuery.new
query.add_query opts[:qobj], :must if opts[:qobj]
query.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless opts[:load_spam] || labels.include?(:spam)
query.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless opts[:load_deleted] || labels.include?(:deleted)
- query.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not unless opts[:load_killed] || labels.include?(:killed)
+ query.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if opts[:skip_killed]
query
end
- def load_sources fn=Redwood::SOURCE_FN
- @sources = Hash[*(Redwood::load_yaml_obj(fn) || []).map { |s| [s.id, s] }.flatten]
- @sources_dirty = false
- end
-
def save_sources fn=Redwood::SOURCE_FN
if @sources_dirty || @sources.any? { |id, s| s.dirty? }
bakfn = fn + ".bak"
if File.exists? fn
File.chmod 0600, fn
- FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(bakfn) > File.size(fn)
+ FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(fn) == 0
end
- Redwood::save_yaml_obj @sources.values, fn
+ Redwood::save_yaml_obj @sources.values.sort_by { |s| s.id.to_i }, fn, true
File.chmod 0600, fn
end
@sources_dirty = false
projects / sup / blobdiff