Currently, graceful exit is a complicated subsystem that keeps a queue
of all pieces expected to be on a node, and asks the node to transfer
those pieces to other nodes one by one. The complexity of the system
has, unfortunately, led to numerous bugs and unexpected behaviors.
We have decided to remove this entire subsystem and restructure graceful
exit as follows:
* Nodes will signal their intent to exit gracefully
* The satellite will not send any new pieces to gracefully exiting nodes
* Pieces on gracefully exiting nodes will be considered by the repair
subsystem as "retrievable but unhealthy". They will be repaired off of
the exiting node as needed.
* After one month (with an appropriately high online score), the node
will be considered exited, and held amounts for the node will be
released. The repair worker will continue to fetch pieces from the
node as long as the node stays online.
* If, at the end of the month, a node's online score is below a certain
threshold, its graceful exit will fail.
Refs: https://github.com/storj/storj/issues/6042
Change-Id: I52d4e07a4198e9cb2adf5e6cee2cb64d6f9f426b
With this change we are removing code responsible for deleting objects
and supporting server side copies created with references. In practice
we are restoring delete queries that we had before server side copy
implementation (with small exception, see bellow).
From deletion queries we are also removing parts with segment metadata
as result because we are not longer sending explicit delete requests to
storage nodes.
https://github.com/storj/storj/issues/5891
Change-Id: Iee4e7a9688cff27a60fb95d60dd233d996f41c85
Test case to verify if server-side copy doesn't affect
GE in any negative way.
Fixes https://github.com/storj/storj/issues/4699
Change-Id: I8c385767cca61499d46d9cb8de7318c56e5d7397
Remove the logic associated to the old transfer queue.
A new transfer queue (gracefulexit_segment_transfer_queue) has been created for migration to segmentLoop.
Transfers from the old queue were not moved to the new queue.
Instead, it was still used for nodes which have initiated graceful exit before migration.
There is no such node left, so we can remove all this logic.
In a next step, we will drop the table.
Change-Id: I3aa9bc29b76065d34b57a73f6e9c9d0297587f54
For being able to use the segment metainfo loop, graceful exit transfers have to include the segment stream_id/position instead of the path. For this, we created a new table graceful_exit_segment_transfer_queue that will replace the graceful_exit_transfer_queue. The table has been created in a previous migration and made accessible through graceful exit db in another one.
This changes makes graceful exit enqueue transfer items for new exiting nodes in the new table.
Change-Id: I7bd00de13e749be521d63ef3b80c168df66b9433
For being able to use the segment metainfo loop, graceful exit transfers have to include the segment stream_id/position instead of the path. For this, we created a new table graceful_exit_segment_transfer_queue that will replace the graceful_exit_transfer_queue. The table has been created in a previous migration.
This change gives access to this table.
Graceful Exit doesn't use the table yet, this will be done in a next change.
Change-Id: I6c09cff4cc45f0529813a8898ddb2d14aadb2cb8
Use the 'AS OF SYSTEM TIME' Cockroach DB clause for the Graceful Exit
(a.k.a GE) queries that count the delete the GE queue items of nodes
which have already exited the network.
Split the subquery used for deleting all the transfer queue items of
nodes which has exited when CRDB is used and batch the queries because
CRDB struggles when executing in a single query unlike Postgres.
The new test which has been added to this commit to verify the CRDB
batch logic for deleting all the transfer queue items of the exited
nodes has raised that the Enqueue method has to run in baches when CRDB
is used otherwise CRDB has return the error "driver: bad connection"
when a big a amount of items are passed to be enqueued. This error
didn't happen with the current test implementation it was with an
initial one that it was creating a big amount of exited nodes and
transfer queue items for those nodes.
Change-Id: I6a099cdbc515a240596bc93141fea3182c2e50a9
The test function fails randomly in the CI when runs with CRDB. There
isn't currently an explanation why the expectation of number of nodes
which exited 4 minutes ago reports 4 nodes rather than 5 and the only
clue that we have now to see if it gets remedied is to give 2 minutes
rather than 1 to the node that exited close to the time passed function
which makes the test to randomly fail.
Change-Id: I3a731e3eb7f19caebdf29713150727f2cf3e0e0a
metabase has become a central concept and it's more suitable for it to
be directly nested under satellite rather than being part of metainfo.
metainfo is going to be the "endpoint" logic for handling requests.
Change-Id: I53770d6761ac1e9a1283b5aa68f471b21e784198
The new 'consistency ge-cleanup-orphaned-data' cli command deleted
orphaned transfer queue items, but not entries in the
graceful_exit_progress table. This will delete orphaned entries
from the exit progress table too.
Change-Id: I5f927aac1f258490678deaf179be92ccfe10fcd8
Using random piece num may generate the exact same key or piecenum.
Instead use fixed piecenum and key.
Change-Id: I54b7bc1a6698149bf99608dd46501ea963cec084
Testing interfaces is slightly clearer when it's in the package needing
the database rather than each individual implementation.
Change-Id: I10334c214a205f7e510b939b4359a2214c4e060a
