summaryrefslogtreecommitdiff
path: root/python/pgq/cascade
AgeCommit message (Collapse)Author
2014-10-08pgq.cascade.admin: fixed takeover ("cannot subscribe to itself")martinko
2014-08-07cascadeadmin: allow drop-node of rootMarko Kreen
2014-01-23pgq.cascade.admin: do not allow subscribing to itself (same node)martinko
2013-04-18londiste status: make node name easier to selectMarko Kreen
2013-04-02cascadeadmin: downgrade missing host= error.Marko Kreen
It's not 100% certain that user does not want such connect strings, so erroring out is not right.
2013-03-22DBScript.add_connect_string_profile: single place for profile logicMarko Kreen
2013-03-22status: forgot to drop cstr_extra argumentMarko Kreen
2013-03-22cascadeadmin: cleanup of profile hack in status commandMarko Kreen
2013-03-22londiste: connection profilesMarko Kreen
Way to add extra connect string parameters for non-config connection strings (public node locations).
2013-03-22cascadeadmin: forgot to make node_name optional everywhereMarko Kreen
2013-03-20cascadeadmin: avoid hardwired db nameMarko Kreen
2013-03-20londiste create: load node name and initial provider also from configMarko Kreen
2013-03-20adminscripts: make info commands not use pidfileMarko Kreen
otherwise they conflict with change operations.
2013-03-06python: text fixesmartinko
2013-03-03cascadeadmin status: wait for threadsMarko Kreen
Instead Queue.join(), call Thread.join() for each thread. Otherwise the thread can be active during process shutdown and throw weird errors. After that, calling Queue.join() seems useless; it can even allow additional failure situations. Also clarify thread count logic.
2013-03-01create-node: validate, accept public location from configMarko Kreen
2013-02-27pgq.cascade.admin: amended threadingmartinko
2013-02-27thread safety catchupmartinko
2013-02-27pgq/cascade: significant speed up of getting status on big cascadesmartinko
e.g.: 1000 node cascade status went from minutes to seconds
2013-02-22noisemartinko
2013-02-22pgq: sweeping change to postpone log string formattingmartinko
2013-02-13noisemartinko
2013-01-09CascadedConsumer is now also subclass of BaseConsumer (does not allow retry)Petr Jelinek
2012-12-04white noisemartinko
2012-11-28Merge remote-tracking branch 'sasha/master'Marko Kreen
- node-status
2012-11-28node-status cleanupSasha Aliashkevich
2012-11-27londiste takeover: fix bug in loading events from other nodeMarko Kreen
code forgot to remember the new most recent tick id
2012-11-26Londiste: node-status command for local nodeSasha Aliashkevich
2012-11-26takeover --dead-root: search for latest events on nodesMarko Kreen
In case of several subscribers to root node, one of them may be ahead of others. Search over cascade for most-ahead subscriber and load events to new root.
2012-11-26londiste resurrect: Get dead root back into cascadeMarko Kreen
In some cases the 'takeover --dead-root' will be used although actual data on node is fine. Eg. network downtime, lost power. It would be good to have a way to bring it back to cascade without need for full rebuild. This command sync queue contents and re-subscribes node. Any events found on old root that are not spread to rest of cascade are dumped and deleted. No actual table changes are made, thus there must be external mechanism to avoid data conflicts.
2012-11-22cascadeadmin: clean help textMarko Kreen
2012-10-18CascadeAdmin: document wait-provider/wait-root commandsMarko Kreen
2012-10-09--sync-watermark: don't allow subtree wm get above upstream wmMarko Kreen
In case root is lagging, the subtree should not drop events.
2012-07-13CascadedWorker: don't move watermark on source further than incoming batch.Marko Kreen
This can happen on merge-leaf-on-branch, where incomong queue can be behind the combined-queue on branch.
2012-07-05londiste: Wait commands for londisteMarko Kreen
add-table --wait-sync Don't return until all pending tables are synced wait-sync Wait until all pending tables are synced. wait-root Wait until root's next tick has appeared locally. wait-provider Wait until provider's next tick has appeared locally.
2012-07-05drop-node: move find_root before local dropMarko Kreen
also put them into separate try/except blocks. This should increase success rate, hopefully. Suggested-By: Eero Oja
2012-06-05CascadedConsumer: re-initialize provider connection when location changesMarko Kreen
This will avoid situation when connection is recreated with old connect string.
2012-05-29CascadedWorker: merge-leaf-to-branch needs to publish wm infoMarko Kreen
Somehow the functionality got lost. Seems the area is in need of some cleanups to make code more clear.
2012-05-25CascadeAdmin: add node name prefix messages from dbMarko Kreen
2012-05-10CascadedWorker: publish_local_wm() is missing a commitMarko Kreen
2012-02-03Drop all txid install code.Marko Kreen
It's obsolete since 8.3. Drop it.
2012-01-31--sync-watermark: limit watermark to specific nodesMarko Kreen
This will allow specified nodes to sync global watermark with each other, without sending it upwards. This isolates root node from lag on downstream nodes.
2011-11-21drop-node: better behaviour when node is downMarko Kreen
2011-11-15change-provider: catch error when unregisteringMarko Kreen
2011-11-14change-provider: move old-provider ops to the endMarko Kreen
That means even if they fail, all the other nodes are already updated.
2011-10-04CascadedConsumer: consistent access to provider dbMarko Kreen
2011-09-07CascadedConsumer: allow --reset without consumer infoMarko Kreen
Can happen for worker on root node.
2011-08-07cascaded unregister-locationEgon Valdmees
Instead of running drop_node function on every node in cascade, unregister-location event added to queue by master
2011-07-04CascadeAdmin: node-info commandMarko Kreen
Shows detailed info about local node. Currently the info needs to be gathered from different places, which is annoying.
2011-06-07leaf branch now accepts add-table eventsEgon Valdmees