[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16417211#comment-16417211 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- [~slebresne] taking this one over, if you don't mind strongly. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Assignee: Aleksey Yeschenko >Priority: Minor > Labels: CommunityFeedbackRequested > Fix For: 4.0, 3.11.x > > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16198459#comment-16198459 ] Sylvain Lebresne commented on CASSANDRA-13910: -- [~iamaleksey]: I'll try to write a patch before eow. I'll ping you though if it looks like this is not going to happen. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Assignee: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > Fix For: 4.0, 3.11.x > > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16196909#comment-16196909 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- bq. Just because you've got plans doesn't mean you should rush removal of a feature that's existed for years. It's not 'just because'. There is two pages of comments here justifying the change, including one in the comment you are replying to. If something has a negative value staying in (which I and others are arguing is the fact), then it should absolutely be removed as soon as possible - which is the next major. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > Fix For: 4.0, 3.11.x > > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16196835#comment-16196835 ] Kurt Greaves commented on CASSANDRA-13910: -- Says me. It's probably the least complex code in the whole read path. Even if you kept it through a re-write it would amount to almost nothing. But that's all besides the point. Just because you've got plans doesn't mean you should rush removal of a feature that's existed for years. It's a database. Change doesn't have to happen at lightspeed. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > Fix For: 4.0, 3.11.x > > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16196820#comment-16196820 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- Right. But you can't *rely* on speculative RR unless your chance is set to an absurdly high value. Which I don't see why anyone would do, and in that case you might as well set speculative retry to {{ALWAYS}} instead. bq. There's also very minimal gain from removing this from the codebase. Says who? I've got plans to rewrite our coordinator read path level. To me having a new clean implementation that doesn't need to concern itself with baggage like RR has more than minimal gain - I only need to worry about speculative retry being a complication. Anyway, repeating my +1 here for removal in 4.0. [~slebresne] Do you want to write up a patch for this? If not, feel free to assign the JIRA to me. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16196782#comment-16196782 ] Robert Stupp commented on CASSANDRA-13910: -- (If nodes are down for longer than the hint-window (or some similar operational issues), you should run an AE repair and not rely on RR.) Considering the misunderstandings in the wild how RR works, I'm also +1 on deprecating it in 3.11.x and removing in 4.0. In fact, I haven't seen a single use case where RR was used intentionally, but many because of "it's the default, so it must be good to have it". > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16196421#comment-16196421 ] Kurt Greaves commented on CASSANDRA-13910: -- I think the proposed timeline for removal is a bit quick. I think if people do consider it a good idea to go ahead with this something like deprecate now, change default to disable in 4.0, and remove in 4.1 if no complaints are had would be far more suitable. Considering people take a long time to upgrade, if we remove in 4.0 users probably won't even realise until 4.1 is already out. I suggest the longer timeline because I do believe there are people with legitimate use cases out there for this. There's also very minimal gain from removing this from the codebase, so there isn't much point in rushing it. Disabling by default is reasonable in the interim. .bq If you do reads at QUORUM, inconsistencies will eventually get read-repaired all the same. Just a tiny bit less quickly. There are a lot of users out there who don't use QUORUM, so this can't really be used as an alternative. Also hints are better but they are still prone to issues. They obviously don't work 100% of the time because I've seen read repairs on clusters that haven't had a node unavailable for longer than the HH window. Hints can also be dropped or get corrupted. Anyway, the main point is there could be people relying on it and we should probably give them at least a few releases to either complain or work out an alternative. On another note, and correct me if I'm wrong, but RR chance is still applied on QUORUM CL's. Without RR triggering I thought that QUORUM would only repair the nodes involved in the QUORUM, rather than all replicas (this is why you need to use ALL if you're trying to repair using read repair). This could be another use case or potentially a suprise. Albeit, not a good use case, but a use case regardless. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore,
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16194604#comment-16194604 ] Carlos Rolo commented on CASSANDRA-13910: - More times that I disable these setting than I left them unchanged. Good with deprecating! > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16190165#comment-16190165 ] Jason Brown commented on CASSANDRA-13910: - bq. I did sent an email to the user list a few days ago here. No feedback on that threads just yet but happy to leave it at least 1-2 more weeks before making any move. Excellent, thank you. I think you can wait up to a week, and if no one is screaming, please proceed full speed ahead. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16190122#comment-16190122 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- bq. C* is successful enough that anything that has been in the product for some amount of time is probably relied upon by someone, somewhere,, for some definition of "rely". Guaranteeing that it's not the case as a bar for removing anything would amount to remove nothing, and that would, imo, be dangerous for the project. Well said. bq. Short of that though, I suggest we move ahead with this rather than keep something we agree is more harmful than helpful most of the time (and we seem to more or less agree on that) on the off chance this may piss off somebody somewhere. +1 > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16189416#comment-16189416 ] Sylvain Lebresne commented on CASSANDRA-13910: -- bq. +1 to creating a thread on @dev / @user for feed back Not sure everybody saw it, but I did sent an email to the user list a few days ago [here|https://mail-archives.apache.org/mod_mbox/cassandra-user/201709.mbox/%3CCAKkz8Q3h9KENiBrsxzPLJfN9oLdgJago8BOmYMuTFkyOkyJm-Q%40mail.gmail.com%3E]. No feedback on that threads just yet but happy to leave it at least 1-2 more weeks before making any move. bq. I would believe that someone, somewhere is relying on it. C* is successful enough that anything that has been in the product for some amount of time is probably relied upon by someone, somewhere,, for some definition of "rely". Guaranteeing that it's not the case as a bar for removing anything would amount to remove nothing, and that would, imo, be dangerous for the project. But don't get me wrong, I'm totally happy to wait more time to see if we get at least a few genuine real-life reports of cases I haven't though about where it's clear those options provide a good trade-off. Short of that though, I suggest we move ahead with this rather than keep something we agree is more harmful than helpful most of the time (and we seem to more or less agree on that) on the off chance this may piss off somebody somewhere. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing >
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16188753#comment-16188753 ] Ben Bromhead commented on CASSANDRA-13910: -- read_repair_chance has caused more pain that it's solved imho. * Inexperienced users sometimes increase this value to insane values when regular repairs are not working, or they are not doing the right thing with their CL. * Doesn't ever make a significant difference in consistency given most real world workload and query distributions - this generally only ends up triggering on the hottest partitions and as mentioned in the issue description will get hit by read_repair anyway. The only good thing about read_repair_chance is it's something you can turn off on a cluster that is under load pressure. This backed up by my many experiences of turning it off :) +1 to creating a thread on @dev / @user for feed back > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16186643#comment-16186643 ] Jeremiah Jordan commented on CASSANDRA-13910: - bq. Asked differently: why would someone set read_repair_chance to a high value? If someone had a multi-dc setup, wrote with LOCAL_ cls, and then read from that DC, they'd trigger the async repair into the other DC, working around the dc local consistency without paying the latency cost on the app side. Sort of a "I can force hint delivery on read without waiting for it" technique. I don't know how common such a use case would be, but I would believe that someone, somewhere is relying on it. I definitely have done exactly this in the past when hints were WAY less reliable. With the current hints implementation that is pretty reliable, I do not see the need for doing something like this anymore. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16185992#comment-16185992 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- bq. If someone had a multi-dc setup, wrote with LOCAL_ cls, and then read from that DC, they'd trigger the async repair into the other DC, working around the dc local consistency without paying the latency cost on the app side. Sort of a "I can force hint delivery on read without waiting for it" technique. I don't know how common such a use case would be, but I would believe that someone, somewhere is relying on it. Right. But so long as they have hints enabled, then even with LOCAL_* cls they'll get the same effect, but more reliably, on write. No waiting for it. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16185983#comment-16185983 ] Jeff Jirsa commented on CASSANDRA-13910: Been thinking about this a bit more. {quote} I'm having a hard time imagining real cases where that trade-off really make sense. {quote} I think the real cases where the trade off make sense are where your earlier assumption is wrong : {quote} I suspect almost everyone use a low "chance" for those options at best (because the extra resources consumption is real), so at the end of the day, it's up to chance how much faster this fixes inconsistencies. {quote} Asked differently: why would someone set {{read_repair_chance}} to a high value? If someone had a multi-dc setup, wrote with {{LOCAL_}} cls, and then read from that DC, they'd trigger the async repair into the other DC, working around the dc local consistency without paying the latency cost on the app side. Sort of a "I can force hint delivery on read without waiting for it" technique. I don't know how common such a use case would be, but I would believe that someone, somewhere is relying on it. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail:
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16185759#comment-16185759 ] Stefan Podkowinski commented on CASSANDRA-13910: I'd also suggest to get some more user feedback on this, before dumping a working, but potentially confusing feature. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > Labels: CommunityFeedbackRequested > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16184549#comment-16184549 ] Jason Brown commented on CASSANDRA-13910: - I like this idea. However, I think it would be good to send out a discussion thread to dev@/user@ to get user input rather than just forcing this change on them. While I find it unlikely, there might be some users who find (or think they find) value from this feature. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16184539#comment-16184539 ] Jeff Jirsa commented on CASSANDRA-13910: {quote} ditto {quote} Second ditto. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16183813#comment-16183813 ] Sam Tunnicliffe commented on CASSANDRA-13910: - bq. Would +1 deprecating it in 3.11.x and removing in 4.0 ditto > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16182600#comment-16182600 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- Also RR still occasionally going beyond local DC for local quorums, if global rr chance is non-zero. I'm not sure we disabled it back then - I think we just changed global rr chance to 0.0 by default and called it a day. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org
[jira] [Commented] (CASSANDRA-13910) Consider deprecating (then removing) read_repair_chance/dclocal_read_repair_chance
[ https://issues.apache.org/jira/browse/CASSANDRA-13910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16182595#comment-16182595 ] Aleksey Yeschenko commented on CASSANDRA-13910: --- There is also some overlap with speculative retries here - and I think our {{ARE}} doesn't handle the case of RR+SR ideally. But I agree. I too find it hard to imagine a scenario when this is noticeably useful. Would +1 deprecating it in 3.11.x and removing in 4.0. > Consider deprecating (then removing) > read_repair_chance/dclocal_read_repair_chance > -- > > Key: CASSANDRA-13910 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13910 > Project: Cassandra > Issue Type: Improvement >Reporter: Sylvain Lebresne >Priority: Minor > > First, let me clarify so this is not misunderstood that I'm not *at all* > suggesting to remove the read-repair mechanism of detecting and repairing > inconsistencies between read responses: that mechanism is imo fine and > useful. But the {{read_repair_chance}} and {{dclocal_read_repair_chance}} > have never been about _enabling_ that mechanism, they are about querying all > replicas (even when this is not required by the consistency level) for the > sole purpose of maybe read-repairing some of the replica that wouldn't have > been queried otherwise. Which btw, bring me to reason 1 for considering their > removal: their naming/behavior is super confusing. Over the years, I've seen > countless users (and not only newbies) misunderstanding what those options > do, and as a consequence misunderstand when read-repair itself was happening. > But my 2nd reason for suggesting this is that I suspect > {{read_repair_chance}}/{{dclocal_read_repair_chance}} are, especially > nowadays, more harmful than anything else when enabled. When those option > kick in, what you trade-off is additional resources consumption (all nodes > have to execute the read) for a _fairly remote chance_ of having some > inconsistencies repaired on _some_ replica _a bit faster_ than they would > otherwise be. To justify that last part, let's recall that: > # most inconsistencies are actually fixed by hints in practice; and in the > case where a node stay dead for a long time so that hints ends up timing-out, > you really should repair the node when it comes back (if not simply > re-bootstrapping it). Read-repair probably don't fix _that_ much stuff in > the first place. > # again, read-repair do happen without those options kicking in. If you do > reads at {{QUORUM}}, inconsistencies will eventually get read-repaired all > the same. Just a tiny bit less quickly. > # I suspect almost everyone use a low "chance" for those options at best > (because the extra resources consumption is real), so at the end of the day, > it's up to chance how much faster this fixes inconsistencies. > Overall, I'm having a hard time imagining real cases where that trade-off > really make sense. Don't get me wrong, those options had their places a long > time ago when hints weren't working all that well, but I think they bring > more confusion than benefits now. > And I think it's sane to reconsider stuffs every once in a while, and to > clean up anything that may not make all that much sense anymore, which I > think is the case here. > Tl;dr, I feel the benefits brought by those options are very slim at best and > well overshadowed by the confusion they bring, and not worth maintaining the > code that supports them (which, to be fair, isn't huge, but getting rid of > {{ReadCallback.AsyncRepairRunner}} wouldn't hurt for instance). > Lastly, if the consensus here ends up being that they can have their use in > weird case and that we fill supporting those cases is worth confusing > everyone else and maintaining that code, I would still suggest disabling them > totally by default. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org